Upload folder using huggingface_hub
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +13 -0
- CLAUDE.md +191 -0
- Report/GRN_CCFM_group_meeting.md +416 -0
- Report/GRN_CCFM_group_meeting.pptx +3 -0
- Report/Nano/.claude/settings.local.json +7 -0
- Report/Nano/Nanomotor_Presentation.pptx +3 -0
- Report/Nano/glucose-fueled-gated-nanomotors-enhancing-in-vivo-anticancer-efficacy-via-deep-drug-penetration-into-tumors.pdf +3 -0
- Report/PPT/GRN_CCFM_presentation.pdf +3 -0
- Report/PPT/GRN_CCFM_presentation.pptx +3 -0
- Report/PPT/create_ppt.js +854 -0
- Report/PPT/fixed-01.jpg +3 -0
- Report/PPT/fixed14-14.jpg +3 -0
- Report/PPT/fixed6-06.jpg +3 -0
- Report/PPT/fixed8-08.jpg +3 -0
- Report/PPT/slide-01.jpg +3 -0
- Report/PPT/slide-02.jpg +3 -0
- Report/PPT/slide-03.jpg +3 -0
- Report/PPT/slide-04.jpg +3 -0
- Report/PPT/slide-05.jpg +3 -0
- Report/PPT/slide-06.jpg +3 -0
- Report/PPT/slide-07.jpg +3 -0
- Report/PPT/slide-08.jpg +3 -0
- Report/PPT/slide-09.jpg +3 -0
- Report/PPT/slide-10.jpg +3 -0
- Report/PPT/slide-11.jpg +3 -0
- Report/PPT/slide-12.jpg +3 -0
- Report/PPT/slide-13.jpg +3 -0
- Report/PPT/slide-14.jpg +3 -0
- Report/PPT/讲解稿.md +183 -0
- Report/PPT2/GRN_CCFM_presentation.pdf +3 -0
- Report/PPT2/GRN_CCFM_presentation.pptx +3 -0
- Report/PPT2/generate.js +781 -0
- Report/PPT2/slide-01.jpg +3 -0
- Report/PPT2/slide-02.jpg +3 -0
- Report/PPT2/slide-03.jpg +3 -0
- Report/PPT2/slide-04.jpg +3 -0
- Report/PPT2/slide-05.jpg +3 -0
- Report/PPT2/slide-06.jpg +3 -0
- Report/PPT2/slide-07.jpg +3 -0
- Report/PPT2/slide-08.jpg +3 -0
- Report/PPT2/slide-09.jpg +3 -0
- Report/PPT2/slide-10.jpg +3 -0
- Report/PPT2/slide-11.jpg +3 -0
- Report/PPT2/slide-12.jpg +3 -0
- Report/PPT2/slide-13.jpg +3 -0
- Report/PPT2/slide-14.jpg +3 -0
- Report/build_pptx.js +703 -0
- Report/week10/GRN_Progress_Report.pdf +3 -0
- Report/week10/GRN_Progress_Report.pptx +3 -0
- Report/week10/gen_pptx.js +874 -0
.gitattributes
CHANGED
|
@@ -136,3 +136,16 @@ GRN/result/scalar/scalar_bs48/eval_only/pred.h5ad filter=lfs diff=lfs merge=lfs
|
|
| 136 |
GRN/result/scalar/scalar_bs48/eval_only/real.h5ad filter=lfs diff=lfs merge=lfs -text
|
| 137 |
GRN/result/topk30_emb/grn-norman-f1-topk30-negTrue-d512-lr5e-05-lw1.0-lp0.4-ema0.9999-ln-wu2000-rk4-cached_sparse-sparse_tk30_L11/iteration_0/pred.h5ad filter=lfs diff=lfs merge=lfs -text
|
| 138 |
GRN/result/topk30_emb/grn-norman-f1-topk30-negTrue-d512-lr5e-05-lw1.0-lp0.4-ema0.9999-ln-wu2000-rk4-cached_sparse-sparse_tk30_L11/iteration_0/real.h5ad filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
GRN/result/scalar/scalar_bs48/eval_only/real.h5ad filter=lfs diff=lfs merge=lfs -text
|
| 137 |
GRN/result/topk30_emb/grn-norman-f1-topk30-negTrue-d512-lr5e-05-lw1.0-lp0.4-ema0.9999-ln-wu2000-rk4-cached_sparse-sparse_tk30_L11/iteration_0/pred.h5ad filter=lfs diff=lfs merge=lfs -text
|
| 138 |
GRN/result/topk30_emb/grn-norman-f1-topk30-negTrue-d512-lr5e-05-lw1.0-lp0.4-ema0.9999-ln-wu2000-rk4-cached_sparse-sparse_tk30_L11/iteration_0/real.h5ad filter=lfs diff=lfs merge=lfs -text
|
| 139 |
+
Report/GRN_CCFM_group_meeting.pptx filter=lfs diff=lfs merge=lfs -text
|
| 140 |
+
Report/Nano/Nanomotor_Presentation.pptx filter=lfs diff=lfs merge=lfs -text
|
| 141 |
+
Report/Nano/glucose-fueled-gated-nanomotors-enhancing-in-vivo-anticancer-efficacy-via-deep-drug-penetration-into-tumors.pdf filter=lfs diff=lfs merge=lfs -text
|
| 142 |
+
Report/PPT/GRN_CCFM_presentation.pdf filter=lfs diff=lfs merge=lfs -text
|
| 143 |
+
Report/PPT/GRN_CCFM_presentation.pptx filter=lfs diff=lfs merge=lfs -text
|
| 144 |
+
Report/PPT2/GRN_CCFM_presentation.pdf filter=lfs diff=lfs merge=lfs -text
|
| 145 |
+
Report/PPT2/GRN_CCFM_presentation.pptx filter=lfs diff=lfs merge=lfs -text
|
| 146 |
+
Report/week10/GRN_Progress_Report.pdf filter=lfs diff=lfs merge=lfs -text
|
| 147 |
+
Report/week10/GRN_Progress_Report.pptx filter=lfs diff=lfs merge=lfs -text
|
| 148 |
+
data/Causal[[:space:]]Schrödinger[[:space:]]Bridges.pdf filter=lfs diff=lfs merge=lfs -text
|
| 149 |
+
data/bridge.pdf filter=lfs diff=lfs merge=lfs -text
|
| 150 |
+
local_bin/soffice filter=lfs diff=lfs merge=lfs -text
|
| 151 |
+
poster/poster/scprompt_poster.pptx filter=lfs diff=lfs merge=lfs -text
|
CLAUDE.md
ADDED
|
@@ -0,0 +1,191 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# CLAUDE.md
|
| 2 |
+
|
| 3 |
+
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
|
| 4 |
+
|
| 5 |
+
## Repository Overview
|
| 6 |
+
|
| 7 |
+
Multi-project AI research repository for single-cell biology and video understanding. All project code lives under `transfer/code/`, data under `transfer/data/`, and the shared Python 3.13 venv in `stack_env/`.
|
| 8 |
+
|
| 9 |
+
### Primary Projects
|
| 10 |
+
|
| 11 |
+
1. **Stack** (`transfer/code/stack/`) — Large-scale encoder-decoder foundation model for single-cell biology (in-context learning on 150M cells). Package: `arc-stack`.
|
| 12 |
+
2. **cell-eval** (`transfer/code/cell-eval/`) — Evaluation metrics suite for single-cell perturbation prediction models. Package: `cell-eval`.
|
| 13 |
+
3. **FOCUS** (`transfer/code/FOCUS/`) — Training-free keyframe selection for long video understanding using multi-armed bandits (ICLR 2026).
|
| 14 |
+
|
| 15 |
+
### Secondary Projects
|
| 16 |
+
|
| 17 |
+
4. **scGPT** (`transfer/code/scGPT/`) — Foundation model for single-cell multi-omics. Uses Poetry.
|
| 18 |
+
5. **scDFM** (`transfer/code/scDFM/`) — Distributional flow matching for single-cell perturbation prediction (ICLR 2026). Uses Conda.
|
| 19 |
+
6. **ori_scDFM** (`transfer/code/ori_scDFM/`) — Original upstream scDFM (cloned from AI4Science-WestlakeU/scDFM). Uses dedicated venv `ori_scDFM_env/` (Python 3.11).
|
| 20 |
+
8. **LatentForcing** (`transfer/code/LatentForcing/`) — Image generation with reordered diffusion trajectories (arXiv 2602.11401).
|
| 21 |
+
9. **CCFM** (`transfer/code/CCFM/`) — Cascaded Conditioned Flow Matching: hybrid of scDFM + LatentForcing + scGPT for guided perturbation prediction. In development.
|
| 22 |
+
10. **adaptive_prompt_selection** (`transfer/code/adaptive_prompt_selection/`) — Bandit-based prompt selection for Stack in-context learning.
|
| 23 |
+
11. **prompt_selection** (`transfer/code/prompt_selection/`) — Evaluation framework and baselines for prompt selection experiments.
|
| 24 |
+
|
| 25 |
+
## HPC Computing Rules (GENKAI Supercomputer)
|
| 26 |
+
|
| 27 |
+
- **NEVER run ML/DL/LLM model inference or training on the login node.** Always submit to compute nodes via `pjsub`.
|
| 28 |
+
- Login node (genkai0002) is only for: editing code, lightweight file operations, `pip install`, job submission, checking results.
|
| 29 |
+
- Lightweight evaluation scripts (e.g., cell-eval metrics, statistical analysis) are acceptable on the login node.
|
| 30 |
+
|
| 31 |
+
### Job Submission (PJM)
|
| 32 |
+
|
| 33 |
+
```bash
|
| 34 |
+
pjsub script.sh # Batch job
|
| 35 |
+
pjsub --interact -L rscgrp=b-inter -L gpu=1 -L elapse=1:00:00 # Interactive GPU (1 GPU, 1h)
|
| 36 |
+
pjstat # Check status
|
| 37 |
+
pjdel <jobid> # Cancel job
|
| 38 |
+
```
|
| 39 |
+
|
| 40 |
+
See `transfer/gpu_batch.sh` and `transfer/gpu_interactive.sh` for job script templates.
|
| 41 |
+
|
| 42 |
+
## Environment & Installation
|
| 43 |
+
|
| 44 |
+
Most projects share `stack_env/` (Python 3.13). Always activate before work:
|
| 45 |
+
|
| 46 |
+
```bash
|
| 47 |
+
source /home/hp250092/ku50001222/qian/aivc/lfj/stack_env/bin/activate
|
| 48 |
+
```
|
| 49 |
+
|
| 50 |
+
**ori_scDFM** uses its own dedicated venv `ori_scDFM_env/` (Python 3.11):
|
| 51 |
+
|
| 52 |
+
```bash
|
| 53 |
+
source /home/hp250092/ku50001222/qian/aivc/lfj/ori_scDFM_env/bin/activate
|
| 54 |
+
```
|
| 55 |
+
|
| 56 |
+
### Installing Projects
|
| 57 |
+
|
| 58 |
+
```bash
|
| 59 |
+
cd transfer/code/stack && pip install -e . # Entry points: stack-train, stack-finetune, stack-embedding, stack-generation
|
| 60 |
+
cd transfer/code/cell-eval && pip install -e . # Entry point: cell-eval (subcommands: prep, run, baseline, score)
|
| 61 |
+
cd transfer/code/FOCUS && pip install -r requirements.txt
|
| 62 |
+
cd transfer/code/scGPT && poetry install
|
| 63 |
+
cd transfer/code/LatentForcing && pip install -r requirements.txt
|
| 64 |
+
# scDFM uses its own conda env: conda env create -f transfer/code/scDFM/environment.yml
|
| 65 |
+
# ori_scDFM uses ori_scDFM_env/ (already installed from environment.yml pip deps)
|
| 66 |
+
# CCFM bootstraps from scDFM: cd transfer/code/CCFM && python _bootstrap_scdfm.py
|
| 67 |
+
```
|
| 68 |
+
|
| 69 |
+
Note: `uv` is available for fast dependency management (used by cell-eval CI).
|
| 70 |
+
|
| 71 |
+
## Common Commands
|
| 72 |
+
|
| 73 |
+
### Stack
|
| 74 |
+
|
| 75 |
+
```bash
|
| 76 |
+
stack-train --config configs/training/bc_large.yaml
|
| 77 |
+
stack-finetune --config configs/finetuning/ft_parsecg.yaml
|
| 78 |
+
stack-embedding --checkpoint <ckpt> --adata <h5ad> --genelist <pkl> --output <out.h5ad>
|
| 79 |
+
stack-generation --checkpoint <ckpt> --base-adata <h5ad> --test-adata <h5ad> --genelist <pkl> --output-dir <dir>
|
| 80 |
+
```
|
| 81 |
+
|
| 82 |
+
### cell-eval
|
| 83 |
+
|
| 84 |
+
```bash
|
| 85 |
+
cell-eval run -ap pred.h5ad -ar real.h5ad --num-threads 64 --profile full
|
| 86 |
+
cell-eval score --user-input agg_results.csv --base-input base_agg_results.csv
|
| 87 |
+
```
|
| 88 |
+
|
| 89 |
+
### FOCUS
|
| 90 |
+
|
| 91 |
+
```bash
|
| 92 |
+
cd transfer/code/FOCUS
|
| 93 |
+
python select_keyframe.py --dataset_name longvideobench --dataset_path <path> --output_dir <dir> --num_keyframes 64
|
| 94 |
+
```
|
| 95 |
+
|
| 96 |
+
### scDFM
|
| 97 |
+
|
| 98 |
+
```bash
|
| 99 |
+
# Uses its own conda env
|
| 100 |
+
cd transfer/code/scDFM
|
| 101 |
+
python src/script/run.py --data norman --batch_size 48 --model_type origin --d_model 128
|
| 102 |
+
```
|
| 103 |
+
|
| 104 |
+
### LatentForcing
|
| 105 |
+
|
| 106 |
+
```bash
|
| 107 |
+
cd transfer/code/LatentForcing
|
| 108 |
+
torchrun --nproc_per_node=8 main_jit.py # Multi-GPU training
|
| 109 |
+
```
|
| 110 |
+
|
| 111 |
+
### Testing & Linting
|
| 112 |
+
|
| 113 |
+
```bash
|
| 114 |
+
# Stack (from transfer/code/stack/)
|
| 115 |
+
pytest tests/
|
| 116 |
+
pytest tests/test_model_core.py -k "test_name" # Single test
|
| 117 |
+
|
| 118 |
+
# cell-eval (from transfer/code/cell-eval/)
|
| 119 |
+
pytest tests/
|
| 120 |
+
ruff check . # Linting (rules: E, F, ERA; max-line-length=120)
|
| 121 |
+
ruff format --check . # Format check (used in CI)
|
| 122 |
+
```
|
| 123 |
+
|
| 124 |
+
## Architecture
|
| 125 |
+
|
| 126 |
+
### Stack (`transfer/code/stack/src/stack/`)
|
| 127 |
+
|
| 128 |
+
Core model uses **Tabular Attention** — alternating cell-wise and gene-wise attention on cell-by-gene matrix chunks:
|
| 129 |
+
|
| 130 |
+
- `models/core/base.py` — `StateICLModelBase`: gene reduction -> positional embedding -> N x `TabularAttentionLayer` -> output MLP. Losses: reconstruction + Sliced Wasserstein distance.
|
| 131 |
+
- `models/core/` — `StateICLModel` (alias: `scShiftAttentionModel`) wraps the base with masking and loss computation.
|
| 132 |
+
- `models/finetune/` — `ICL_FinetunedModel` with frozen-teacher distillation via `LightningFinetunedModel`.
|
| 133 |
+
- `modules/attention.py` — `MultiHeadAttention`, `TabularAttentionLayer` (cell-attn + gene-attn per layer).
|
| 134 |
+
- `modules/regularizers.py` — `SlicedWassersteinDistance`.
|
| 135 |
+
- `training/` — `LightningGeneModel` (PyTorch Lightning wrapper), `DataModule`, scheduler utils.
|
| 136 |
+
- `finetune/` — `LightningFinetunedModel` (student-teacher EMA), finetuning `DataModule`.
|
| 137 |
+
- `data/` — Dataset configs, HVG computation, H5 data management, sparse matrix loading.
|
| 138 |
+
- `cli/` — Entry points: `launch_training`, `launch_finetuning`, `embedding`, `generation`.
|
| 139 |
+
- Config: YAML files in `configs/training/` and `configs/finetuning/`. CLI args override config values.
|
| 140 |
+
|
| 141 |
+
### cell-eval (`transfer/code/cell-eval/src/cell_eval/`)
|
| 142 |
+
|
| 143 |
+
Uses the **registry pattern** for metrics:
|
| 144 |
+
|
| 145 |
+
- `_evaluator.py` — `MetricsEvaluator`: takes predicted/real AnnData, runs DE (via `pdex`), computes metrics.
|
| 146 |
+
- `metrics/_registry.py` — `MetricRegistry` with `register()` / `compute()`. Metrics are AnnData-based or DE-based.
|
| 147 |
+
- `metrics/_impl.py`, `_de.py`, `_anndata.py` — Metric implementations.
|
| 148 |
+
- `_pipeline/` — `MetricPipeline` with named profiles (e.g., `full`).
|
| 149 |
+
- `_score.py` — Baseline-normalized scoring.
|
| 150 |
+
- `_cli/` — Subcommands: `prep`, `run`, `baseline`, `score`.
|
| 151 |
+
- `_types/` — Typed containers: `PerturbationAnndataPair`, `DEComparison`.
|
| 152 |
+
|
| 153 |
+
### FOCUS (`transfer/code/FOCUS/`)
|
| 154 |
+
|
| 155 |
+
Two-file architecture:
|
| 156 |
+
- `focus.py` — `FOCUS` class: pure CPE bandit algorithm (no I/O). Coarse exploration -> fine exploitation using Bernstein confidence bounds.
|
| 157 |
+
- `select_keyframe.py` — Data pipeline: video loading (decord), BLIP similarity scoring (LAVIS), result output.
|
| 158 |
+
|
| 159 |
+
### scDFM (`transfer/code/scDFM/src/`)
|
| 160 |
+
|
| 161 |
+
Flow matching for perturbation prediction: `flow_matching/` (algorithm), `models/` (networks), `tokenizer/` (cell/gene tokens), `loss/` (custom losses), `data_process/` (loading).
|
| 162 |
+
|
| 163 |
+
### LatentForcing (`transfer/code/LatentForcing/`)
|
| 164 |
+
|
| 165 |
+
Diffusion with reordered trajectories. Model variants: `model_jit.py`, `model_cot.py`, `model_repa.py`. Training engine: `engine_jit.py`. Entry: `main_jit.py`.
|
| 166 |
+
|
| 167 |
+
### CCFM (`transfer/code/CCFM/`)
|
| 168 |
+
|
| 169 |
+
Cascaded flow matching combining scDFM + LatentForcing denoiser + scGPT embeddings. Entry: `scripts/run_cascaded.py`. Config: `config/config_cascaded.py`. Imports scDFM modules via `_scdfm_imports.py` bridge.
|
| 170 |
+
|
| 171 |
+
### adaptive_prompt_selection (`transfer/code/adaptive_prompt_selection/`)
|
| 172 |
+
|
| 173 |
+
Bandit-based selection of in-context examples for Stack. `adaptive_prompt.py` (orchestrator), `cell_bandit.py` (bandit algorithm). Run via `run_experiment.py`.
|
| 174 |
+
|
| 175 |
+
## Dataset Details
|
| 176 |
+
|
| 177 |
+
Main dataset (`transfer/data/stack_train/20260203_Parse_10M_PBMC_cytokines.h5ad`): 9.7M cells x 40K genes, 12 donors, 91 cytokines, 18 cell types. Key obs columns: `donor`, `cytokine`, `treatment`, `cell_type`, `sample`. Stack dataset config format: `"human:parse_bio:sample:cell_type:false"`.
|
| 178 |
+
|
| 179 |
+
All single-cell projects use **AnnData** (`.h5ad`) as the standard data format.
|
| 180 |
+
|
| 181 |
+
## CI/CD
|
| 182 |
+
|
| 183 |
+
- **cell-eval**: GitHub Actions CI (`uv sync`, `ruff format --check`, `pytest -v`, CLI smoke test) on push/PR. Python 3.12.
|
| 184 |
+
- **Stack**: GitHub Actions publishes to PyPI on release (trusted publishing via setuptools build).
|
| 185 |
+
- **scGPT**: Claude Code integration workflow for PR review.
|
| 186 |
+
|
| 187 |
+
## GPU Job Notes
|
| 188 |
+
|
| 189 |
+
- Stack batch jobs set `PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:256` to avoid CUDA OOM fragmentation.
|
| 190 |
+
- Job templates: `transfer/gpu_batch.sh` (batch, 1 GPU, 3h), `transfer/gpu_interactive.sh` (interactive, configurable hours/GPUs).
|
| 191 |
+
- Login node and compute nodes share the same filesystem path: `/home/hp250092/ku50001222/qian/aivc/lfj/`.
|
Report/GRN_CCFM_group_meeting.md
ADDED
|
@@ -0,0 +1,416 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GRN-Guided Cascaded Flow Matching for Single-Cell Perturbation Prediction
|
| 2 |
+
|
| 3 |
+
## 组会汇报
|
| 4 |
+
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
## 1. Task:单细胞扰动预测
|
| 8 |
+
|
| 9 |
+
### 虚拟细胞(Virtual Cell)
|
| 10 |
+
|
| 11 |
+
**虚拟细胞**是当前计算生物学的核心愿景:构建一个能在计算机中模拟真实细胞行为的AI模型——给定任意输入条件(基因型、环境、扰动),预测细胞的分子状态变化。单细胞扰动预测是实现虚拟细胞最关键的子任务之一。
|
| 12 |
+
|
| 13 |
+
### 什么是扰动?
|
| 14 |
+
|
| 15 |
+
扰动(Perturbation)是指人为改变细胞的某些条件,观察细胞如何响应。常见的扰动类型包括:
|
| 16 |
+
|
| 17 |
+
- **药物扰动(Drug perturbation)**:施加小分子化合物或药物,观察细胞的转录组变化(如 L1000/LINCS 数据)
|
| 18 |
+
- **细胞因子扰动(Cytokine perturbation)**:用细胞因子(如 IL-6、TNF-α、IFN-γ 等)刺激细胞,研究免疫信号通路响应
|
| 19 |
+
- **基因扰动(Genetic perturbation)**:直接改变基因的功能状态,包括:
|
| 20 |
+
- **基因敲除(Knock-out, KO)**:用 CRISPR-Cas9 完全删除目标基因的功能
|
| 21 |
+
- **基因过表达(Overexpression, OE)**:人为提高目标基因的表达量(如 CRISPRa 激活)
|
| 22 |
+
- **基因敲低(Knock-down, KD)**:用 RNAi/shRNA 部分降低目标基因的表达(不完全删除)
|
| 23 |
+
|
| 24 |
+
本工作聚焦**基因扰动**,使用 Perturb-seq 技术:对细胞施加 CRISPR 基因扰动,然后用单细胞 RNA-seq 测量每个细胞所有基因的表达变化。
|
| 25 |
+
|
| 26 |
+
### 任务形式化
|
| 27 |
+
|
| 28 |
+
```
|
| 29 |
+
已知: x_ctrl ∈ R^G (control 细胞的基因表达谱, G ≈ 5000 高变基因)
|
| 30 |
+
p ∈ {gene₁, gene₂} (被扰动的基因)
|
| 31 |
+
|
| 32 |
+
预测: x_pert ∈ R^G (扰动后细胞的基因表达谱)
|
| 33 |
+
```
|
| 34 |
+
|
| 35 |
+
### 为什么重要?
|
| 36 |
+
|
| 37 |
+
- **药物筛选加速**:wet-lab 做一次 Perturb-seq 实验成本极高($$$);如果能计算预测,可以大幅缩小候选范围
|
| 38 |
+
- **组合扰动爆炸**:N 个基因的两两组合 = N(N-1)/2 种实验,不可能穷举,必须靠预测
|
| 39 |
+
- **理解疾病机制**:预测哪些基因被扰动后会产生某种疾病表型
|
| 40 |
+
|
| 41 |
+
### 数据特点
|
| 42 |
+
|
| 43 |
+
- Norman et al. 数据集:~9K 细胞 × 5000 HVG,105 种单/双基因 CRISPR 扰动(KO + OE)
|
| 44 |
+
- **细胞配对不可得**:我们无法得到 (x_ctrl_i, x_pert_i) 的逐细胞配对——扰动是破坏性的,一个细胞只能测一次
|
| 45 |
+
- 评估:DE 基因 overlap、方向一致性、MSE、Pearson 相关等
|
| 46 |
+
|
| 47 |
+
---
|
| 48 |
+
|
| 49 |
+
## 2. 现有方法综述与批判
|
| 50 |
+
|
| 51 |
+
### 2.1 简单基线
|
| 52 |
+
|
| 53 |
+
**Additive shift(均值偏移)**
|
| 54 |
+
|
| 55 |
+
```
|
| 56 |
+
x̂_pert = x_ctrl + mean(x_pert_train - x_ctrl_train)
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
- 假设:扰动效应对所有细胞是**常数平移**
|
| 60 |
+
- 问题:完全忽略细胞异质性;同一扰动对不同细胞类型的效应可能截然不同
|
| 61 |
+
- 但**出奇地难以超越**——很多复杂模型在 top DE 基因上并不比它好
|
| 62 |
+
|
| 63 |
+
### 2.2 基于预训练大模型的方法
|
| 64 |
+
|
| 65 |
+
**scGPT** (Nature Methods 2024)
|
| 66 |
+
|
| 67 |
+
- 自回归 transformer,在大规模单细胞数据上预训练
|
| 68 |
+
- 扰动预测方式:fine-tune,将扰动基因 token mask 掉,让模型"补全"
|
| 69 |
+
- **问题**:
|
| 70 |
+
- 本质是**自回归补全**,不是为扰动预测设计的目标函数
|
| 71 |
+
- 编码的是细胞的**绝对状态**,不直接建模状态变化
|
| 72 |
+
- Fine-tuning 过拟合风险高(训练样本少)
|
| 73 |
+
|
| 74 |
+
**Geneformer** (Nature 2024)
|
| 75 |
+
|
| 76 |
+
- 基于 rank-value encoding 的 transformer 预训练
|
| 77 |
+
- 用 in-silico perturbation:直接删除目标基因 token,看 embedding 变化
|
| 78 |
+
- **问题**:
|
| 79 |
+
- in-silico perturbation 是**启发式**的,没有学习扰动的动力学
|
| 80 |
+
- rank encoding 丢失表达量信息
|
| 81 |
+
- 不能处理组合扰动
|
| 82 |
+
|
| 83 |
+
### 2.3 专用扰动预测模型
|
| 84 |
+
|
| 85 |
+
**CPA** (Molecular Systems Biology 2023)
|
| 86 |
+
|
| 87 |
+
- Compositional Perturbation Autoencoder:将细胞状态分解为 basal state + perturbation effect + covariate effect
|
| 88 |
+
- 用加法假设在 latent space 组合
|
| 89 |
+
- **问题**:
|
| 90 |
+
- **线性可加假设**过强:基因调控是非线性的
|
| 91 |
+
- Autoencoder 重建质量限制上限
|
| 92 |
+
- 不建模基因间的调控关系
|
| 93 |
+
|
| 94 |
+
**GEARS** (Nature Biotechnology 2023)
|
| 95 |
+
|
| 96 |
+
- 用 Gene Ontology (GO) 图上的 GNN 编码基因关系
|
| 97 |
+
- Cross-attention 融合扰动信息和细胞状态
|
| 98 |
+
- **问题**:
|
| 99 |
+
- GO 图是**静态先验知识**,不随细胞状态变化
|
| 100 |
+
- 图结构的选择对结果影响很大(GO vs PPI vs GRN)
|
| 101 |
+
- 仍然是确定性预测,不能给出分布
|
| 102 |
+
|
| 103 |
+
**STATE** (ICLR 2025)
|
| 104 |
+
|
| 105 |
+
- Stacked Attention for Expression Transformation
|
| 106 |
+
- **问题**:
|
| 107 |
+
- 确定性预测
|
| 108 |
+
- 仍然没有从 GRN 变化角度建模
|
| 109 |
+
|
| 110 |
+
**CellFlow** (preprint 2025)
|
| 111 |
+
|
| 112 |
+
- 也是 flow matching 框架
|
| 113 |
+
- 用预训练 embedding 作为条件
|
| 114 |
+
- **问题**:
|
| 115 |
+
- 预训练 embedding 仍然编码绝对状态
|
| 116 |
+
- 没有显式建模扰动对调控网络的改变
|
| 117 |
+
|
| 118 |
+
**scDFM** (ICLR 2026)
|
| 119 |
+
|
| 120 |
+
- 将 **Conditional Flow Matching** 引入扰动预测
|
| 121 |
+
- 学习从噪声到 target 表达的速度场,条件为 control 表达 + 扰动 ID
|
| 122 |
+
- DiffPerceiverBlock:差分注意力 + cross-attention
|
| 123 |
+
- **优点**���生成式模型,能给出分布;flow matching 训练稳定
|
| 124 |
+
- **问题**:
|
| 125 |
+
- **信息来源单一**:只有 control 的数值表达 + 扰动基因的 embedding
|
| 126 |
+
- **不理解基因调控关系**:模型需要从数据中隐式学习 GRN,而训练数据极少
|
| 127 |
+
- d_model=128,表达能力有限
|
| 128 |
+
|
| 129 |
+
### 2.4 共性问题总结
|
| 130 |
+
|
| 131 |
+
```
|
| 132 |
+
所有现有方法的共同盲区:
|
| 133 |
+
|
| 134 |
+
扰动 → [黑箱模型] → 表达变化
|
| 135 |
+
|
| 136 |
+
没有任何一个方法显式地建模:
|
| 137 |
+
|
| 138 |
+
扰动 → 基因调控网络变化 → 表达变化
|
| 139 |
+
↑
|
| 140 |
+
这一步被跳过了
|
| 141 |
+
```
|
| 142 |
+
|
| 143 |
+
---
|
| 144 |
+
|
| 145 |
+
## 3. 我们的 Motivation
|
| 146 |
+
|
| 147 |
+
### 3.1 Motivation 1:流匹配解决细胞配对问题
|
| 148 |
+
|
| 149 |
+
单细胞扰动预测的根本困难:**没有 paired data**。
|
| 150 |
+
|
| 151 |
+
我们无法得到 (x_ctrl_i, x_pert_i) 这样的配对——一个细胞扰动后就变了,不能回到扰动前状态。
|
| 152 |
+
|
| 153 |
+
传统方法的处理:群体均值匹配(丢失异质性)或 autoencoder(受限于重建质量)。
|
| 154 |
+
|
| 155 |
+
**Flow Matching 的优势**:它学习的是从 source 分布到 target 分布的**概率传输映射**,天然适合 unpaired 数据:
|
| 156 |
+
|
| 157 |
+
```
|
| 158 |
+
p(x_ctrl) ──(学习 ODE 路径)──→ p(x_pert | perturbation)
|
| 159 |
+
```
|
| 160 |
+
|
| 161 |
+
- 不需要逐细胞配对,只需要两组细胞的**群体分布**
|
| 162 |
+
- 通过 Conditional Optimal Transport 构造训练对,比随机配对效率高
|
| 163 |
+
- 生成式输出:每个 control 细胞可以采样多个 prediction,给出**不确定性估计**
|
| 164 |
+
|
| 165 |
+
### 3.2 Motivation 2:从 GRN 变化角度理解扰动
|
| 166 |
+
|
| 167 |
+
**关键生物学事实**:基因扰动不是简单地改变一个基因的值。
|
| 168 |
+
|
| 169 |
+
```
|
| 170 |
+
CRISPR knock-out 基因 A
|
| 171 |
+
↓
|
| 172 |
+
基因 A 的表达降为 0
|
| 173 |
+
↓
|
| 174 |
+
A 直接调控的基因 B, C, D 表达改变(一级效应)
|
| 175 |
+
↓
|
| 176 |
+
B 调控的 E, F,C 调控的 G, H ... 依次改变(级联效应)
|
| 177 |
+
↓
|
| 178 |
+
最终数千个基因的表达都发生变化
|
| 179 |
+
```
|
| 180 |
+
|
| 181 |
+
这个级联传播的路径,就是 **基因调控网络 (GRN)**。
|
| 182 |
+
|
| 183 |
+
**核心洞察**:如果我们能先理解扰动如何改变了 GRN,再基于改变后的 GRN 预测表达,预测会更准确。
|
| 184 |
+
|
| 185 |
+
```
|
| 186 |
+
现有方法: 扰动 ───────────────────────→ 表达变化
|
| 187 |
+
(端到端黑箱)
|
| 188 |
+
|
| 189 |
+
我们: 扰动 → GRN 变化 (Δ_attn) → 表达变化
|
| 190 |
+
↑ 显式建模 ↑ 有据可依
|
| 191 |
+
```
|
| 192 |
+
|
| 193 |
+
### 3.3 Motivation 3:scGPT 的 Attention ≈ 数据驱动的 GRN
|
| 194 |
+
|
| 195 |
+
预训练的 scGPT 模型在大规模单细胞数据上学到了基因间的调控关系,这些关系被编码在 transformer 的 **attention matrix** 中:
|
| 196 |
+
|
| 197 |
+
```
|
| 198 |
+
attn[i][j] 高 → 基因 j 的状态对基因 i 的表达有强影响
|
| 199 |
+
```
|
| 200 |
+
|
| 201 |
+
这等价于一个**数据驱动的、上下文相关的 GRN**——它随细胞状态而变化,比静态 GO 图更灵活。
|
| 202 |
+
|
| 203 |
+
我们可以用**同一模型**,分别输入 control 和 perturbation 的表达,得到两个 attention matrix,它们的差直接反映**扰动引起的 GRN 变化**。
|
| 204 |
+
|
| 205 |
+
---
|
| 206 |
+
|
| 207 |
+
## 4. 我们的方法
|
| 208 |
+
|
| 209 |
+
### 4.1 总体思路
|
| 210 |
+
|
| 211 |
+
在 scDFM 的 flow matching 框架上,引入 **GRN-aware latent flow**:
|
| 212 |
+
|
| 213 |
+
```
|
| 214 |
+
Cascaded Flow Matching:
|
| 215 |
+
┌─────────────────────────────────┐ ┌──────────────────────────────┐
|
| 216 |
+
│ GRN Latent Flow │ │ Expression Flow │
|
| 217 |
+
│ │ │ │
|
| 218 |
+
│ noise → GRN 变化特征 │ │ noise → 基因表达 │
|
| 219 |
+
│ (理解调控网络如何改变) │ │ (基于 GRN 变化预测表达) │
|
| 220 |
+
│ │ │ │
|
| 221 |
+
│ 推理: 先完成 │ → │ 推理: 后完成 │
|
| 222 |
+
└─────────────────────────────────┘ └──────────────────────────────┘
|
| 223 |
+
Stage 1 Stage 2
|
| 224 |
+
```
|
| 225 |
+
|
| 226 |
+
**直觉**:模型先"想清楚"扰动改变了哪些基因调控关系,再基于这些理解去预测表达变化。
|
| 227 |
+
|
| 228 |
+
### 4.2 GRN 特征:Attention-Delta
|
| 229 |
+
|
| 230 |
+
```
|
| 231 |
+
Δ_attn = Attention(perturbed cell) - Attention(control cell)
|
| 232 |
+
|
| 233 |
+
GRN features = Δ_attn @ gene_embeddings
|
| 234 |
+
```
|
| 235 |
+
|
| 236 |
+
- `Δ_attn` 捕捉扰动引起的调控关系变化
|
| 237 |
+
- `@ gene_embeddings` 将稀疏的调控变化信号聚合到每个基因的连续特征空间
|
| 238 |
+
- 输出:每个基因一个 512 维向量,编码"这个基因上游的调控关系发生了怎样的变化"
|
| 239 |
+
|
| 240 |
+
### 4.3 模型架构
|
| 241 |
+
|
| 242 |
+
```
|
| 243 |
+
条件信息 (推理时可用) 辅助生成目标 (推理时从噪声生成)
|
| 244 |
+
┌──────────────────────────┐ ┌───────────────────────────────┐
|
| 245 |
+
│ control 表达 x_ctrl │ │ GRN 变化特征 z (512-d/gene) │
|
| 246 |
+
│ 扰动基因 ID pert_id │ │ = Δ_attn @ gene_emb │
|
| 247 |
+
│ 时间步 t₁, t₂ │ │ 来自 frozen scGPT │
|
| 248 |
+
└──────────────────────────┘ └───────────────────────────────┘
|
| 249 |
+
↓ ↓
|
| 250 |
+
┌─── Expression Stream ───┐ ┌─── Latent Stream ───┐
|
| 251 |
+
│ GeneEncoder + ValueEnc │ │ LatentEmbedder │
|
| 252 |
+
│ → expr_tokens (B,G,d) │ │ → lat_tokens(B,G,d)│
|
| 253 |
+
└─────────────┬───────────┘ └──────────┬──────────┘
|
| 254 |
+
└───────── (+) ─────────────┘
|
| 255 |
+
↓
|
| 256 |
+
x = expr + latent (加法融合)
|
| 257 |
+
↓
|
| 258 |
+
┌─── Conditioning ───┐
|
| 259 |
+
│ c = t₁ + t₂ + pert │
|
| 260 |
+
└────────┬──────────┘
|
| 261 |
+
↓
|
| 262 |
+
┌─── Shared Backbone ──┐
|
| 263 |
+
│ DiffPerceiverBlock │ × 4 layers
|
| 264 |
+
│ (with GeneadaLN) │
|
| 265 |
+
└────────┬────────────┘
|
| 266 |
+
┌────┴────┐
|
| 267 |
+
┌────┴───┐ ┌──┴────────┐
|
| 268 |
+
│ExprHead │ │LatentHead │
|
| 269 |
+
│→ v_expr │ │→ v_latent │
|
| 270 |
+
│ (B,G) │ │(B,G,512) │
|
| 271 |
+
└─────────┘ └───────────┘
|
| 272 |
+
```
|
| 273 |
+
|
| 274 |
+
### 4.4 Cascaded 训练
|
| 275 |
+
|
| 276 |
+
训练时不同时优化两个 flow,而是**概率切换**:
|
| 277 |
+
|
| 278 |
+
```
|
| 279 |
+
随机 coin flip:
|
| 280 |
+
40% → 训练 Latent Flow: t₂ 随机, t₁ = 0, 只算 loss_latent
|
| 281 |
+
60% → 训练 Expression Flow: t₁ 随机, t₂ ≈ 1, 只算 loss_expr
|
| 282 |
+
```
|
| 283 |
+
|
| 284 |
+
### 4.5 Cascaded 推理(核心)
|
| 285 |
+
|
| 286 |
+
```
|
| 287 |
+
Stage 1: z_noise ═══(ODE)═══> z_clean (GRN latent: 0→1)
|
| 288 |
+
此时 expression 静止 (t₁=0)
|
| 289 |
+
|
| 290 |
+
Stage 2: x_noise ═══(ODE)═══> x_pred (Expression: 0→1)
|
| 291 |
+
此时 GRN latent 已完成 (t₂=1)
|
| 292 |
+
```
|
| 293 |
+
|
| 294 |
+
**生物学直觉**:先理解调控网络如何变化,再基于变化后的调控网络预测表达。
|
| 295 |
+
|
| 296 |
+
### 4.6 架构图绘制 Prompt (Nano Banana)
|
| 297 |
+
|
| 298 |
+
> **Prompt for architecture diagram:**
|
| 299 |
+
>
|
| 300 |
+
> Create a clean, professional scientific figure illustrating the "GRN-Guided Cascaded Flow Matching" architecture for single-cell perturbation prediction. Use a horizontal layout, left-to-right flow, with a soft blue-white color scheme.
|
| 301 |
+
>
|
| 302 |
+
> **Left panel — "GRN Feature Extraction" (frozen, no gradient):**
|
| 303 |
+
> Show a frozen scGPT transformer icon (with a snowflake symbol). Two inputs feed into it: "control expression x_ctrl" (blue arrow) and "perturbed expression x_pert" (red arrow). Each produces an attention matrix (shown as a small heatmap grid). Between the two heatmaps, show a minus sign, producing "Δ_attn" (a difference heatmap). Then show Δ_attn multiplied by "Gene Embeddings" (a matrix icon), producing "GRN change features z" as a (G × 512) colored matrix strip. Label this output with dimension annotations (B, G, 512). Add a dashed box around the entire left panel labeled "Frozen scGPT — GRN Change Extraction".
|
| 304 |
+
>
|
| 305 |
+
> **Center panel — "Cascaded Flow Model" (trainable):**
|
| 306 |
+
> Show two parallel input streams at the top:
|
| 307 |
+
> (1) "Expression Stream": x_t (noised target expression) + x_ctrl (control) → two small encoder blocks → fusion → "expr_tokens"
|
| 308 |
+
> (2) "Latent Stream": z_t (noised GRN features) → LatentEmbedder → "latent_tokens"
|
| 309 |
+
> Show these two streams merging with a "+" symbol into combined tokens.
|
| 310 |
+
> Below, show a conditioning vector "c = t_expr + t_latent + pert_emb" feeding into a vertical stack of 4 shared backbone blocks (labeled "DiffPerceiverBlock × 4").
|
| 311 |
+
> At the bottom, the backbone splits into two decoder heads: "Expression Head → v_expr (B, G)" on the left, and "Latent Head → v_latent (B, G, 512)" on the right.
|
| 312 |
+
>
|
| 313 |
+
> **Right panel — "Cascaded Inference":**
|
| 314 |
+
> Show a two-stage timeline diagram:
|
| 315 |
+
> Stage 1 (top): "GRN Latent Flow" — an ODE trajectory from z_noise (random dots) to z_clean (structured pattern), with t_latent going from 0 to 1. Label: "First: understand how GRN changes".
|
| 316 |
+
> Stage 2 (bottom): "Expression Flow" — an ODE trajectory from x_noise to x_pred (a gene expression barplot), with t_expr going from 0 to 1. Label: "Then: predict expression changes". Show a downward arrow from Stage 1 to Stage 2 indicating "z_clean conditions Stage 2".
|
| 317 |
+
>
|
| 318 |
+
> **Bottom banner:** A biological intuition strip showing: "Perturbation → GRN rewiring (Δ attention) → Gene expression change", with small icons (a gene network graph changing, then expression bars changing).
|
| 319 |
+
>
|
| 320 |
+
> Style: Nature/Cell journal style, vector graphics, minimal clutter, no 3D effects, consistent font sizes, use color coding (blue for expression path, orange/gold for latent/GRN path, gray for frozen components).
|
| 321 |
+
|
| 322 |
+
---
|
| 323 |
+
|
| 324 |
+
## 5. 当前问题与解决方向
|
| 325 |
+
|
| 326 |
+
### 5.1 问题:GRN 信号提取困难——噪声淹没真信号
|
| 327 |
+
|
| 328 |
+
scGPT 的 attention matrix 是一个 (G × G) 的**稠密矩阵**——每对基因之间都有非零 attention 值。
|
| 329 |
+
|
| 330 |
+
但真实的基因调控网络是**极度稀疏的**:一个基因通常只直接调控几十个下游基因,而不是几千个。
|
| 331 |
+
|
| 332 |
+
```
|
| 333 |
+
Attention matrix: 5000 × 5000 = 25,000,000 个非零值
|
| 334 |
+
真实 GRN: 每个基因平均 ~20-50 个调控靶标
|
| 335 |
+
|
| 336 |
+
→ 99%+ 的 attention 值是噪声,不是真正的调控关系
|
| 337 |
+
```
|
| 338 |
+
|
| 339 |
+
这意味着 `Δ_attn @ gene_emb` 里绝大部分信号是噪声的变化,真正的 GRN 变化信号被淹没。
|
| 340 |
+
|
| 341 |
+
**当前观察佐证**:latent loss ≈ 1.12,远高于 expression loss ≈ 0.019,说明模型难以预测 latent 速度场——因为目标本身就充满噪声。
|
| 342 |
+
|
| 343 |
+
**解决方向——稀疏化 top-K**:
|
| 344 |
+
|
| 345 |
+
```
|
| 346 |
+
原始: Δ_attn (G × G) 全部 25M 值 → 噪声多,信号弱
|
| 347 |
+
↓ 稀疏化
|
| 348 |
+
Top-K: 每个基因只保留 |Δ| 最大的 K=30 个 → 过滤 99.4% 噪声
|
| 349 |
+
↓
|
| 350 |
+
features = sparse_Δ_topk @ gene_emb
|
| 351 |
+
```
|
| 352 |
+
|
| 353 |
+
已实现并在实验中(`sparse_topk_emb` 模式)。
|
| 354 |
+
|
| 355 |
+
### 5.2 问题:512 维 latent 预测难度过大
|
| 356 |
+
|
| 357 |
+
每个基因的 GRN 特征是 512 维向量,整个 latent target 是 (G, 512) = 250 万维的速度场——模型需要在每个时间步预测这么大的向量。
|
| 358 |
+
|
| 359 |
+
**消融实验证实**:将 latent 维度从 512 降到 1,latent loss 从 ~1.1 降到 ~0.5-0.7——维度是难度的重要来源。
|
| 360 |
+
|
| 361 |
+
**解决方向——PCA 降维**:
|
| 362 |
+
|
| 363 |
+
```
|
| 364 |
+
512-d gene_emb → PCA 投影到 64 维
|
| 365 |
+
features = sparse_Δ_topk @ pca_basis → (B, G, 64)
|
| 366 |
+
```
|
| 367 |
+
|
| 368 |
+
用 PCA 找到 attention delta 信号的主要变化方向,将 512 维压缩到 64 维,去掉冗余维度。scgpt_dim=64,bottleneck_dim=64。
|
| 369 |
+
|
| 370 |
+
已实现并在实验中(`sparse_pca` 模式)。
|
| 371 |
+
|
| 372 |
+
---
|
| 373 |
+
|
| 374 |
+
## 6. 总结与展望
|
| 375 |
+
|
| 376 |
+
### 6.1 核心贡献
|
| 377 |
+
|
| 378 |
+
我们的工作**不是在模型架构上做改进**,而是**从生物学机制出发**重新建模扰动预测任务:
|
| 379 |
+
|
| 380 |
+
```
|
| 381 |
+
现有方法: perturbation → [model] → expression change
|
| 382 |
+
↑
|
| 383 |
+
纯数据驱动黑箱
|
| 384 |
+
|
| 385 |
+
我们: perturbation → GRN rewiring → expression change
|
| 386 |
+
↑
|
| 387 |
+
显式建模生物学机制
|
| 388 |
+
```
|
| 389 |
+
|
| 390 |
+
用 Cascaded Flow Matching 的两阶段结构来实现"先理解调控变化,再预测表达变化"。
|
| 391 |
+
|
| 392 |
+
### 6.2 后续关键实验:验证因果假设
|
| 393 |
+
|
| 394 |
+
目前的 cascaded 推理是**固定顺序**的:先 latent(GRN),后 expression。
|
| 395 |
+
|
| 396 |
+
这是否真的重要?我们计划做一个关键实验来验证:
|
| 397 |
+
|
| 398 |
+
**实验设计**:训练时 t₁ 和 t₂ 完全随机采样(不再 cascade),模型支持**任意顺序推理**。然后对比:
|
| 399 |
+
|
| 400 |
+
| 推理方式 | 含义 | 预期 |
|
| 401 |
+
|---------|------|------|
|
| 402 |
+
| 先 GRN latent → 后 expression | 先理解调控变化,再预测表达 | **最优** |
|
| 403 |
+
| 先 expression → 后 GRN latent | 先预测表达,再理解调控 | 次优 |
|
| 404 |
+
| 同时 random | 无显式顺序 | 最差 |
|
| 405 |
+
|
| 406 |
+
如果"先 GRN 后 expression"显著优于其他顺序,就验证了:
|
| 407 |
+
|
| 408 |
+
> **理解基因调控网络的变化,是预测扰动表达变化的前提条件,而不是副产物。**
|
| 409 |
+
|
| 410 |
+
这就是我们这个工作要回答的核心科学问题。
|
| 411 |
+
|
| 412 |
+
---
|
| 413 |
+
|
| 414 |
+
### 一句话总结
|
| 415 |
+
|
| 416 |
+
> 我们用 scGPT 的 attention delta 显式提取扰动引起的基因调控网络变化,通过 cascaded flow matching 强制模型"先理解 GRN 如何改变,再预测表达如何变化",从而将生物学先验——扰动通过 GRN 级联传播——融入生成式模型的推理过程。
|
Report/GRN_CCFM_group_meeting.pptx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:96c5298cbc5a080fcd9619a795f851bc908988d0af0b4de7130ad0aa2f71c80c
|
| 3 |
+
size 304227
|
Report/Nano/.claude/settings.local.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"permissions": {
|
| 3 |
+
"allow": [
|
| 4 |
+
"Bash(npm list:*)"
|
| 5 |
+
]
|
| 6 |
+
}
|
| 7 |
+
}
|
Report/Nano/Nanomotor_Presentation.pptx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0076d5bf4be0040027beb2b575e8ce2a137387a4cd75200b46df65ad8be79b86
|
| 3 |
+
size 260965
|
Report/Nano/glucose-fueled-gated-nanomotors-enhancing-in-vivo-anticancer-efficacy-via-deep-drug-penetration-into-tumors.pdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5dd6f6dd36fb853ba8f62accf2c08e1d1a3dbb9f6c69d01919c262e3c06f563f
|
| 3 |
+
size 5089152
|
Report/PPT/GRN_CCFM_presentation.pdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:853c91d2b5685cc60ae9b2ae249a66b6ce789bfece3fe7d6bc4df1cf3fd6fc62
|
| 3 |
+
size 469947
|
Report/PPT/GRN_CCFM_presentation.pptx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fbbf29a0bf16d5a014e4f58bc7ce9f6609af8fd608b43bdd4b939633ed407cba
|
| 3 |
+
size 317866
|
Report/PPT/create_ppt.js
ADDED
|
@@ -0,0 +1,854 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
const pptxgen = require("pptxgenjs");
|
| 2 |
+
|
| 3 |
+
const pres = new pptxgen();
|
| 4 |
+
pres.layout = "LAYOUT_16x9";
|
| 5 |
+
pres.author = "Qian";
|
| 6 |
+
pres.title = "GRN-Guided Cascaded Flow Matching for Single-Cell Perturbation Prediction";
|
| 7 |
+
|
| 8 |
+
// === COLOR PALETTE: Deep Teal + Amber (biology meets computation) ===
|
| 9 |
+
const C = {
|
| 10 |
+
dark: "0F172A",
|
| 11 |
+
light: "F8FAFC",
|
| 12 |
+
teal: "0891B2",
|
| 13 |
+
tealDk: "164E63",
|
| 14 |
+
tealLt: "ECFEFF",
|
| 15 |
+
white: "FFFFFF",
|
| 16 |
+
text: "1E293B",
|
| 17 |
+
textS: "475569",
|
| 18 |
+
textM: "94A3B8",
|
| 19 |
+
red: "EF4444",
|
| 20 |
+
green: "10B981",
|
| 21 |
+
amber: "F59E0B",
|
| 22 |
+
purple: "8B5CF6",
|
| 23 |
+
blue: "3B82F6",
|
| 24 |
+
border: "E2E8F0",
|
| 25 |
+
};
|
| 26 |
+
|
| 27 |
+
const shadow = () => ({ type: "outer", color: "000000", blur: 6, offset: 2, angle: 135, opacity: 0.1 });
|
| 28 |
+
|
| 29 |
+
// Slide number helper (bottom-right)
|
| 30 |
+
let slideNum = 0;
|
| 31 |
+
function addSlideNum(sl, dark) {
|
| 32 |
+
slideNum++;
|
| 33 |
+
sl.addText(String(slideNum), {
|
| 34 |
+
x: 9.2, y: 5.15, w: 0.5, h: 0.3,
|
| 35 |
+
fontSize: 10, fontFace: "Calibri", color: dark ? C.textM : C.textS, align: "right", margin: 0,
|
| 36 |
+
});
|
| 37 |
+
}
|
| 38 |
+
|
| 39 |
+
function card(sl, x, y, w, h, opts = {}) {
|
| 40 |
+
sl.addShape(pres.shapes.RECTANGLE, {
|
| 41 |
+
x, y, w, h, fill: { color: opts.fill || C.white }, shadow: shadow(),
|
| 42 |
+
});
|
| 43 |
+
if (opts.accent) {
|
| 44 |
+
sl.addShape(pres.shapes.RECTANGLE, { x, y, w: 0.06, h, fill: { color: opts.accent } });
|
| 45 |
+
}
|
| 46 |
+
}
|
| 47 |
+
|
| 48 |
+
function titleBar(sl, title) {
|
| 49 |
+
sl.addShape(pres.shapes.RECTANGLE, { x: 0, y: 0, w: 10, h: 0.9, fill: { color: C.tealDk } });
|
| 50 |
+
sl.addText(title, {
|
| 51 |
+
x: 0.6, y: 0.15, w: 8.8, h: 0.6,
|
| 52 |
+
fontSize: 24, fontFace: "Georgia", color: C.white, bold: true, margin: 0,
|
| 53 |
+
});
|
| 54 |
+
}
|
| 55 |
+
|
| 56 |
+
// ============================
|
| 57 |
+
// SLIDE 1: TITLE
|
| 58 |
+
// ============================
|
| 59 |
+
let s = pres.addSlide();
|
| 60 |
+
s.background = { color: C.dark };
|
| 61 |
+
// Left teal block as visual motif
|
| 62 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0, y: 0, w: 0.15, h: 5.625, fill: { color: C.teal } });
|
| 63 |
+
|
| 64 |
+
s.addText("GRN-Guided Cascaded\nFlow Matching", {
|
| 65 |
+
x: 0.8, y: 1.1, w: 8.4, h: 1.8,
|
| 66 |
+
fontSize: 42, fontFace: "Georgia", color: C.white, bold: true, lineSpacingMultiple: 1.15,
|
| 67 |
+
});
|
| 68 |
+
s.addText("for Single-Cell Perturbation Prediction", {
|
| 69 |
+
x: 0.8, y: 2.95, w: 8.4, h: 0.5,
|
| 70 |
+
fontSize: 20, fontFace: "Calibri", color: "22D3EE",
|
| 71 |
+
});
|
| 72 |
+
s.addText("组会汇报", {
|
| 73 |
+
x: 0.8, y: 4.2, w: 3, h: 0.4,
|
| 74 |
+
fontSize: 14, fontFace: "Calibri", color: C.textM,
|
| 75 |
+
});
|
| 76 |
+
addSlideNum(s, true);
|
| 77 |
+
|
| 78 |
+
// ============================
|
| 79 |
+
// SLIDE 2: TASK
|
| 80 |
+
// ============================
|
| 81 |
+
s = pres.addSlide();
|
| 82 |
+
s.background = { color: C.light };
|
| 83 |
+
titleBar(s, "Task:单细胞扰动预测");
|
| 84 |
+
|
| 85 |
+
// Virtual Cell card
|
| 86 |
+
card(s, 0.5, 1.15, 4.3, 2.1, { accent: C.teal });
|
| 87 |
+
s.addText("虚拟细胞 (Virtual Cell)", {
|
| 88 |
+
x: 0.75, y: 1.25, w: 3.8, h: 0.4,
|
| 89 |
+
fontSize: 16, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 90 |
+
});
|
| 91 |
+
s.addText("构建能模拟真实细胞行为的 AI 模型:给定任意输入条件(基因型、环境、扰动),预测细胞的分子状态变化。单细胞扰动预测是实现虚拟细胞最关键的子任务。", {
|
| 92 |
+
x: 0.75, y: 1.7, w: 3.85, h: 1.4,
|
| 93 |
+
fontSize: 12.5, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.35,
|
| 94 |
+
});
|
| 95 |
+
|
| 96 |
+
// Perturbation types card
|
| 97 |
+
card(s, 5.2, 1.15, 4.3, 2.1, { accent: C.amber });
|
| 98 |
+
s.addText("扰动类型", {
|
| 99 |
+
x: 5.45, y: 1.25, w: 3.8, h: 0.4,
|
| 100 |
+
fontSize: 16, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 101 |
+
});
|
| 102 |
+
s.addText([
|
| 103 |
+
{ text: "药物扰动", options: { bold: true, breakLine: true, fontSize: 12 } },
|
| 104 |
+
{ text: " 小分子化合物 → 转录组变化", options: { breakLine: true, fontSize: 11, color: C.textS } },
|
| 105 |
+
{ text: "细胞因子扰动", options: { bold: true, breakLine: true, fontSize: 12 } },
|
| 106 |
+
{ text: " IL-6, TNF-α 等 → 信号通路响应", options: { breakLine: true, fontSize: 11, color: C.textS } },
|
| 107 |
+
{ text: "基因扰动(本工作聚焦)", options: { bold: true, breakLine: true, fontSize: 12, color: C.teal } },
|
| 108 |
+
{ text: " CRISPR KO / OE / KD → 全基因组变化", options: { fontSize: 11, color: C.textS } },
|
| 109 |
+
], { x: 5.45, y: 1.7, w: 3.85, h: 1.5, margin: 0, lineSpacingMultiple: 1.2 });
|
| 110 |
+
|
| 111 |
+
// Task formalization
|
| 112 |
+
card(s, 0.5, 3.5, 9.0, 1.6, { accent: C.purple });
|
| 113 |
+
s.addText("任务形式化", {
|
| 114 |
+
x: 0.75, y: 3.6, w: 2, h: 0.35,
|
| 115 |
+
fontSize: 14, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 116 |
+
});
|
| 117 |
+
s.addText([
|
| 118 |
+
{ text: "已知: ", options: { bold: true, fontSize: 13 } },
|
| 119 |
+
{ text: "x_ctrl (control 基因表达, G ~ 5000 HVG) + p (扰动基因)", options: { fontSize: 13, breakLine: true } },
|
| 120 |
+
{ text: "预测: ", options: { bold: true, fontSize: 13 } },
|
| 121 |
+
{ text: "x_pert (扰动后基因表达谱)", options: { fontSize: 13 } },
|
| 122 |
+
], { x: 0.75, y: 4.0, w: 8.5, h: 0.9, fontFace: "Consolas", color: C.text, margin: 0, lineSpacingMultiple: 1.5 });
|
| 123 |
+
addSlideNum(s);
|
| 124 |
+
|
| 125 |
+
// ============================
|
| 126 |
+
// SLIDE 3: WHY IMPORTANT + DATA
|
| 127 |
+
// ============================
|
| 128 |
+
s = pres.addSlide();
|
| 129 |
+
s.background = { color: C.light };
|
| 130 |
+
titleBar(s, "为什么重要 & 数据特点");
|
| 131 |
+
|
| 132 |
+
// Three importance cards
|
| 133 |
+
const imp = [
|
| 134 |
+
{ icon: "$$$", title: "药物筛选加速", desc: "Perturb-seq 成本极高\n计算预测大幅缩小候选范围", c: C.teal },
|
| 135 |
+
{ icon: "N\u00B2", title: "组合扰动爆炸", desc: "N 基因两两组合 = N(N-1)/2\n不可能穷举,必须靠预测", c: C.amber },
|
| 136 |
+
{ icon: "\u2697", title: "理解疾病机制", desc: "预测哪些基因扰动\n产生特定疾病表型", c: C.purple },
|
| 137 |
+
];
|
| 138 |
+
imp.forEach((it, i) => {
|
| 139 |
+
const x = 0.5 + i * 3.1;
|
| 140 |
+
card(s, x, 1.15, 2.8, 2.0, { accent: it.c });
|
| 141 |
+
s.addText(it.icon, {
|
| 142 |
+
x: x + 0.2, y: 1.25, w: 2.4, h: 0.55,
|
| 143 |
+
fontSize: 28, fontFace: "Georgia", color: it.c, bold: true, margin: 0,
|
| 144 |
+
});
|
| 145 |
+
s.addText(it.title, {
|
| 146 |
+
x: x + 0.2, y: 1.8, w: 2.4, h: 0.3,
|
| 147 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 148 |
+
});
|
| 149 |
+
s.addText(it.desc, {
|
| 150 |
+
x: x + 0.2, y: 2.15, w: 2.4, h: 0.85,
|
| 151 |
+
fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, lineSpacingMultiple: 1.3,
|
| 152 |
+
});
|
| 153 |
+
});
|
| 154 |
+
|
| 155 |
+
// Data card
|
| 156 |
+
card(s, 0.5, 3.5, 9.0, 1.65, { accent: C.tealDk });
|
| 157 |
+
s.addText("Norman 数据集", {
|
| 158 |
+
x: 0.75, y: 3.6, w: 3, h: 0.35,
|
| 159 |
+
fontSize: 16, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 160 |
+
});
|
| 161 |
+
const stats = [
|
| 162 |
+
{ val: "~9K", label: "细胞数" },
|
| 163 |
+
{ val: "5000", label: "高变基因" },
|
| 164 |
+
{ val: "105", label: "扰动条件" },
|
| 165 |
+
{ val: "KO+OE", label: "扰动类型" },
|
| 166 |
+
];
|
| 167 |
+
stats.forEach((st, i) => {
|
| 168 |
+
const x = 0.75 + i * 2.15;
|
| 169 |
+
s.addText(st.val, {
|
| 170 |
+
x, y: 4.05, w: 1.9, h: 0.5,
|
| 171 |
+
fontSize: 26, fontFace: "Georgia", color: C.teal, bold: true, margin: 0,
|
| 172 |
+
});
|
| 173 |
+
s.addText(st.label, {
|
| 174 |
+
x, y: 4.5, w: 1.9, h: 0.25,
|
| 175 |
+
fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0,
|
| 176 |
+
});
|
| 177 |
+
});
|
| 178 |
+
s.addText("关键挑战:细胞配对不可得 — 扰动是破坏性的,一个细胞只能测一次", {
|
| 179 |
+
x: 0.75, y: 4.85, w: 8.5, h: 0.2,
|
| 180 |
+
fontSize: 11, fontFace: "Calibri", color: C.red, italic: true, margin: 0,
|
| 181 |
+
});
|
| 182 |
+
|
| 183 |
+
// ============================
|
| 184 |
+
addSlideNum(s);
|
| 185 |
+
// SLIDE 4: EXISTING METHODS (1)
|
| 186 |
+
// ============================
|
| 187 |
+
s = pres.addSlide();
|
| 188 |
+
s.background = { color: C.light };
|
| 189 |
+
titleBar(s, "现有方法:简单基线与预训练大模型");
|
| 190 |
+
|
| 191 |
+
// Additive Shift
|
| 192 |
+
card(s, 0.5, 1.15, 4.3, 1.85, { accent: C.textM });
|
| 193 |
+
s.addText("Additive Shift(均值偏移)", {
|
| 194 |
+
x: 0.75, y: 1.25, w: 3.8, h: 0.3,
|
| 195 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 196 |
+
});
|
| 197 |
+
s.addText("x\u0302_pert = x_ctrl + mean(x_pert - x_ctrl)", {
|
| 198 |
+
x: 0.75, y: 1.6, w: 3.8, h: 0.25,
|
| 199 |
+
fontSize: 11, fontFace: "Consolas", color: C.tealDk, margin: 0,
|
| 200 |
+
});
|
| 201 |
+
s.addText([
|
| 202 |
+
{ text: "假设扰动效应对所有细胞是常数平移", options: { breakLine: true } },
|
| 203 |
+
{ text: "忽略细胞异质性", options: { breakLine: true } },
|
| 204 |
+
{ text: "但出奇地难以超越", options: { color: C.red, bold: true } },
|
| 205 |
+
], { x: 0.75, y: 1.95, w: 3.8, h: 0.85, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 206 |
+
|
| 207 |
+
// scGPT
|
| 208 |
+
card(s, 5.2, 1.15, 4.3, 1.85, { accent: C.blue });
|
| 209 |
+
s.addText("scGPT (Nature Methods 2024)", {
|
| 210 |
+
x: 5.45, y: 1.25, w: 3.8, h: 0.3,
|
| 211 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 212 |
+
});
|
| 213 |
+
s.addText([
|
| 214 |
+
{ text: "自回归 Transformer 大规模预训练", options: { breakLine: true } },
|
| 215 |
+
{ text: "扰动基因 mask → 模型补全", options: { breakLine: true } },
|
| 216 |
+
{ text: "本质是自回归补全,非扰动预测目标", options: { breakLine: true, color: C.red } },
|
| 217 |
+
{ text: "编码绝对状态,不建模状态变化", options: { color: C.red } },
|
| 218 |
+
], { x: 5.45, y: 1.7, w: 3.8, h: 1.1, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 219 |
+
|
| 220 |
+
// Geneformer
|
| 221 |
+
card(s, 0.5, 3.25, 4.3, 1.85, { accent: C.green });
|
| 222 |
+
s.addText("Geneformer (Nature 2024)", {
|
| 223 |
+
x: 0.75, y: 3.35, w: 3.8, h: 0.3,
|
| 224 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 225 |
+
});
|
| 226 |
+
s.addText([
|
| 227 |
+
{ text: "Rank-value encoding + Transformer", options: { breakLine: true } },
|
| 228 |
+
{ text: "In-silico perturbation: 删除基因 token", options: { breakLine: true } },
|
| 229 |
+
{ text: "启发式方法,没学习扰动动力学", options: { breakLine: true, color: C.red } },
|
| 230 |
+
{ text: "Rank encoding 丢失表达量信息", options: { color: C.red } },
|
| 231 |
+
], { x: 0.75, y: 3.7, w: 3.8, h: 1.1, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 232 |
+
|
| 233 |
+
// CPA
|
| 234 |
+
card(s, 5.2, 3.25, 4.3, 1.85, { accent: C.amber });
|
| 235 |
+
s.addText("CPA (Mol Sys Bio 2023)", {
|
| 236 |
+
x: 5.45, y: 3.35, w: 3.8, h: 0.3,
|
| 237 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 238 |
+
});
|
| 239 |
+
s.addText([
|
| 240 |
+
{ text: "Basal + perturbation + covariate 可加分解", options: { breakLine: true } },
|
| 241 |
+
{ text: "Latent space 线性组合", options: { breakLine: true } },
|
| 242 |
+
{ text: "线性可加假设过强", options: { breakLine: true, color: C.red } },
|
| 243 |
+
{ text: "不建模基因间调控关系", options: { color: C.red } },
|
| 244 |
+
], { x: 5.45, y: 3.7, w: 3.8, h: 1.1, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 245 |
+
|
| 246 |
+
// ============================
|
| 247 |
+
addSlideNum(s);
|
| 248 |
+
// SLIDE 5: EXISTING METHODS (2)
|
| 249 |
+
// ============================
|
| 250 |
+
s = pres.addSlide();
|
| 251 |
+
s.background = { color: C.light };
|
| 252 |
+
titleBar(s, "现有方法:专用扰动预测模型");
|
| 253 |
+
|
| 254 |
+
// GEARS
|
| 255 |
+
card(s, 0.5, 1.15, 4.3, 1.7, { accent: C.purple });
|
| 256 |
+
s.addText("GEARS (Nat Biotech 2023)", {
|
| 257 |
+
x: 0.75, y: 1.25, w: 3.8, h: 0.3,
|
| 258 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 259 |
+
});
|
| 260 |
+
s.addText([
|
| 261 |
+
{ text: "GO 图 + GNN 编码基因关系", options: { breakLine: true } },
|
| 262 |
+
{ text: "GO 是静态先验,不随细胞状态变", options: { breakLine: true, color: C.red } },
|
| 263 |
+
{ text: "确定性预测,不能给出分布", options: { color: C.red } },
|
| 264 |
+
], { x: 0.75, y: 1.65, w: 3.8, h: 0.95, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 265 |
+
|
| 266 |
+
// STATE
|
| 267 |
+
card(s, 5.2, 1.15, 4.3, 1.7, { accent: C.teal });
|
| 268 |
+
s.addText("STATE (ICLR 2025)", {
|
| 269 |
+
x: 5.45, y: 1.25, w: 3.8, h: 0.3,
|
| 270 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 271 |
+
});
|
| 272 |
+
s.addText([
|
| 273 |
+
{ text: "Stacked Attention for Expression Transform", options: { breakLine: true } },
|
| 274 |
+
{ text: "确定性预测", options: { breakLine: true, color: C.red } },
|
| 275 |
+
{ text: "没有从 GRN 变化角度建模", options: { color: C.red } },
|
| 276 |
+
], { x: 5.45, y: 1.65, w: 3.8, h: 0.95, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 277 |
+
|
| 278 |
+
// CellFlow
|
| 279 |
+
card(s, 0.5, 3.1, 4.3, 1.7, { accent: C.blue });
|
| 280 |
+
s.addText("CellFlow (preprint 2025)", {
|
| 281 |
+
x: 0.75, y: 3.2, w: 3.8, h: 0.3,
|
| 282 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 283 |
+
});
|
| 284 |
+
s.addText([
|
| 285 |
+
{ text: "Flow matching + 预训练 embedding 条件", options: { breakLine: true } },
|
| 286 |
+
{ text: "预训练 embedding 编码绝对状态", options: { breakLine: true, color: C.red } },
|
| 287 |
+
{ text: "没有显式建模调控网络改变", options: { color: C.red } },
|
| 288 |
+
], { x: 0.75, y: 3.6, w: 3.8, h: 0.95, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 289 |
+
|
| 290 |
+
// scDFM
|
| 291 |
+
card(s, 5.2, 3.1, 4.3, 1.7, { accent: C.teal });
|
| 292 |
+
s.addText("scDFM (ICLR 2026)", {
|
| 293 |
+
x: 5.45, y: 3.2, w: 3.8, h: 0.3,
|
| 294 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 295 |
+
});
|
| 296 |
+
s.addText([
|
| 297 |
+
{ text: "Conditional FM + DiffPerceiverBlock", options: { breakLine: true } },
|
| 298 |
+
{ text: "生成式模型,训练稳定", options: { breakLine: true, color: C.green } },
|
| 299 |
+
{ text: "信息来源单一,不理解 GRN", options: { breakLine: true, color: C.red } },
|
| 300 |
+
{ text: "d_model=128,表达能力有限", options: { color: C.red } },
|
| 301 |
+
], { x: 5.45, y: 3.6, w: 3.8, h: 1.0, fontSize: 11, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 302 |
+
|
| 303 |
+
// ============================
|
| 304 |
+
addSlideNum(s);
|
| 305 |
+
// SLIDE 6: BLIND SPOT (section divider)
|
| 306 |
+
// ============================
|
| 307 |
+
s = pres.addSlide();
|
| 308 |
+
s.background = { color: C.tealDk };
|
| 309 |
+
|
| 310 |
+
s.addText("所有现有方法的共同盲区", {
|
| 311 |
+
x: 0.8, y: 0.6, w: 8.4, h: 0.7,
|
| 312 |
+
fontSize: 30, fontFace: "Georgia", color: C.white, bold: true, margin: 0,
|
| 313 |
+
});
|
| 314 |
+
|
| 315 |
+
// Current: black box
|
| 316 |
+
card(s, 0.8, 1.6, 8.4, 1.4, { fill: "1B4A5A" });
|
| 317 |
+
s.addText("现有方法", {
|
| 318 |
+
x: 1.0, y: 1.65, w: 2, h: 0.3,
|
| 319 |
+
fontSize: 12, fontFace: "Calibri", color: "CBD5E1", margin: 0,
|
| 320 |
+
});
|
| 321 |
+
s.addText([
|
| 322 |
+
{ text: "扰动", options: { bold: true, fontSize: 20 } },
|
| 323 |
+
{ text: " \u2192 ", options: { fontSize: 20, color: C.textM } },
|
| 324 |
+
{ text: "[ \u9ED1\u7BB1\u6A21\u578B ]", options: { bold: true, fontSize: 20, color: C.red } },
|
| 325 |
+
{ text: " \u2192 ", options: { fontSize: 20, color: C.textM } },
|
| 326 |
+
{ text: "表达变化", options: { bold: true, fontSize: 20 } },
|
| 327 |
+
], { x: 1.0, y: 2.05, w: 8.0, h: 0.7, fontFace: "Calibri", color: C.white, align: "center", margin: 0 });
|
| 328 |
+
|
| 329 |
+
// Ours: explicit GRN
|
| 330 |
+
card(s, 0.8, 3.3, 8.4, 1.6, { fill: "0C3547" });
|
| 331 |
+
s.addText("我们的方法", {
|
| 332 |
+
x: 1.0, y: 3.35, w: 2, h: 0.3,
|
| 333 |
+
fontSize: 12, fontFace: "Calibri", color: "22D3EE", margin: 0,
|
| 334 |
+
});
|
| 335 |
+
s.addText([
|
| 336 |
+
{ text: "扰动", options: { bold: true, fontSize: 20 } },
|
| 337 |
+
{ text: " \u2192 ", options: { fontSize: 20, color: C.textM } },
|
| 338 |
+
{ text: "GRN 变化", options: { bold: true, fontSize: 20, color: C.teal } },
|
| 339 |
+
{ text: " \u2192 ", options: { fontSize: 20, color: C.textM } },
|
| 340 |
+
{ text: "表达变化", options: { bold: true, fontSize: 20 } },
|
| 341 |
+
], { x: 1.0, y: 3.75, w: 8.0, h: 0.7, fontFace: "Calibri", color: C.white, align: "center", margin: 0 });
|
| 342 |
+
s.addText("\u2191 显式建模生物学机制", {
|
| 343 |
+
x: 3.0, y: 4.4, w: 4, h: 0.3,
|
| 344 |
+
fontSize: 14, fontFace: "Calibri", color: "22D3EE", align: "center", margin: 0,
|
| 345 |
+
});
|
| 346 |
+
|
| 347 |
+
// ============================
|
| 348 |
+
addSlideNum(s, true);
|
| 349 |
+
// SLIDE 7: MOTIVATION 1 - FLOW MATCHING
|
| 350 |
+
// ============================
|
| 351 |
+
s = pres.addSlide();
|
| 352 |
+
s.background = { color: C.light };
|
| 353 |
+
titleBar(s, "Motivation 1:Flow Matching 解决配对问题");
|
| 354 |
+
|
| 355 |
+
// Problem
|
| 356 |
+
card(s, 0.5, 1.15, 4.3, 2.0, { accent: C.red });
|
| 357 |
+
s.addText("核心困难:无 Paired Data", {
|
| 358 |
+
x: 0.75, y: 1.25, w: 3.8, h: 0.3,
|
| 359 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 360 |
+
});
|
| 361 |
+
s.addText([
|
| 362 |
+
{ text: "一个细胞扰动后就变了,无法回到扰动前", options: { breakLine: true } },
|
| 363 |
+
{ text: "无法得到 (x_ctrl_i, x_pert_i) 逐细胞配对", options: { breakLine: true } },
|
| 364 |
+
{ text: "传统方法:群体均值匹配(丢失异质性)", options: { breakLine: true } },
|
| 365 |
+
{ text: "或 Autoencoder(受限于重建质量)", options: {} },
|
| 366 |
+
], { x: 0.75, y: 1.65, w: 3.8, h: 1.3, fontSize: 12, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 367 |
+
|
| 368 |
+
// Solution
|
| 369 |
+
card(s, 5.2, 1.15, 4.3, 2.0, { accent: C.green });
|
| 370 |
+
s.addText("Flow Matching 的优势", {
|
| 371 |
+
x: 5.45, y: 1.25, w: 3.8, h: 0.3,
|
| 372 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 373 |
+
});
|
| 374 |
+
s.addText([
|
| 375 |
+
{ text: "学习分布间的概率传输映射", options: { breakLine: true } },
|
| 376 |
+
{ text: "不需逐细胞配对,只需群体分布", options: { breakLine: true } },
|
| 377 |
+
{ text: "Conditional OT 构造高效训练对", options: { breakLine: true } },
|
| 378 |
+
{ text: "生成式输出 \u2192 不确定性估计", options: {} },
|
| 379 |
+
], { x: 5.45, y: 1.65, w: 3.8, h: 1.3, fontSize: 12, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 380 |
+
|
| 381 |
+
// Flow diagram
|
| 382 |
+
card(s, 0.5, 3.5, 9.0, 1.6);
|
| 383 |
+
s.addText("p(x_ctrl)", {
|
| 384 |
+
x: 0.8, y: 3.8, w: 2.2, h: 0.6,
|
| 385 |
+
fontSize: 18, fontFace: "Consolas", color: C.tealDk, bold: true, align: "center", margin: 0,
|
| 386 |
+
});
|
| 387 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 388 |
+
x: 3.2, y: 4.0, w: 3.6, h: 0.28, fill: { color: C.teal, transparency: 30 },
|
| 389 |
+
});
|
| 390 |
+
s.addText("\u2500\u2500 \u5B66\u4E60 ODE \u8DEF\u5F84 \u2500\u2500\u25B6", {
|
| 391 |
+
x: 3.2, y: 3.95, w: 3.6, h: 0.35,
|
| 392 |
+
fontSize: 13, fontFace: "Calibri", color: C.white, align: "center", margin: 0,
|
| 393 |
+
});
|
| 394 |
+
s.addText("p(x_pert | pert)", {
|
| 395 |
+
x: 7.0, y: 3.8, w: 2.5, h: 0.6,
|
| 396 |
+
fontSize: 18, fontFace: "Consolas", color: C.tealDk, bold: true, align: "center", margin: 0,
|
| 397 |
+
});
|
| 398 |
+
s.addText("天然适合 unpaired data \u2014 只需两组细胞的群体分布即可训练", {
|
| 399 |
+
x: 0.8, y: 4.55, w: 8.4, h: 0.35,
|
| 400 |
+
fontSize: 12, fontFace: "Calibri", color: C.teal, italic: true, margin: 0, align: "center",
|
| 401 |
+
});
|
| 402 |
+
|
| 403 |
+
// ============================
|
| 404 |
+
addSlideNum(s);
|
| 405 |
+
// SLIDE 8: MOTIVATION 2&3 - GRN + ATTENTION
|
| 406 |
+
// ============================
|
| 407 |
+
s = pres.addSlide();
|
| 408 |
+
s.background = { color: C.light };
|
| 409 |
+
titleBar(s, "Motivation 2 & 3:GRN + scGPT Attention");
|
| 410 |
+
|
| 411 |
+
// Left: GRN cascade
|
| 412 |
+
card(s, 0.5, 1.15, 4.3, 4.0, { accent: C.amber });
|
| 413 |
+
s.addText("扰动通过 GRN 级联传播", {
|
| 414 |
+
x: 0.75, y: 1.25, w: 3.8, h: 0.3,
|
| 415 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 416 |
+
});
|
| 417 |
+
s.addText([
|
| 418 |
+
{ text: "CRISPR KO 基因 A", options: { bold: true, breakLine: true, fontSize: 12, color: C.tealDk } },
|
| 419 |
+
{ text: " \u2193", options: { breakLine: true, fontSize: 11, color: C.textM } },
|
| 420 |
+
{ text: "A 的表达降为 0", options: { breakLine: true, fontSize: 12 } },
|
| 421 |
+
{ text: " \u2193", options: { breakLine: true, fontSize: 11, color: C.textM } },
|
| 422 |
+
{ text: "一级效应:A \u2192 B, C, D 改变", options: { bold: true, breakLine: true, fontSize: 12 } },
|
| 423 |
+
{ text: " \u2193", options: { breakLine: true, fontSize: 11, color: C.textM } },
|
| 424 |
+
{ text: "级联效应:B\u2192E,F C\u2192G,H ...", options: { bold: true, breakLine: true, fontSize: 12 } },
|
| 425 |
+
{ text: " \u2193", options: { breakLine: true, fontSize: 11, color: C.textM } },
|
| 426 |
+
{ text: "最终数千个基因表达变化", options: { bold: true, fontSize: 12, color: C.red } },
|
| 427 |
+
], { x: 0.75, y: 1.7, w: 3.8, h: 2.8, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.15 });
|
| 428 |
+
s.addText("核心:先理解 GRN 变化 \u2192 再预测表达", {
|
| 429 |
+
x: 0.75, y: 4.55, w: 3.8, h: 0.35,
|
| 430 |
+
fontSize: 12, fontFace: "Calibri", color: C.teal, bold: true, italic: true, margin: 0,
|
| 431 |
+
});
|
| 432 |
+
|
| 433 |
+
// Right: scGPT Attention
|
| 434 |
+
card(s, 5.2, 1.15, 4.3, 4.0, { accent: C.teal });
|
| 435 |
+
s.addText("scGPT Attention \u2248 数据驱动 GRN", {
|
| 436 |
+
x: 5.45, y: 1.25, w: 3.8, h: 0.3,
|
| 437 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 438 |
+
});
|
| 439 |
+
s.addText("attn[i][j] 高 \u2192 基因 j 对基因 i 有强调控", {
|
| 440 |
+
x: 5.45, y: 1.7, w: 3.8, h: 0.3,
|
| 441 |
+
fontSize: 11, fontFace: "Consolas", color: C.tealDk, margin: 0,
|
| 442 |
+
});
|
| 443 |
+
s.addText([
|
| 444 |
+
{ text: "上下文相关的 GRN", options: { bold: true, breakLine: true, fontSize: 12 } },
|
| 445 |
+
{ text: "随细胞状态变化,比静态 GO 图更灵活", options: { breakLine: true, fontSize: 11, color: C.textS } },
|
| 446 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 447 |
+
{ text: "提取扰动引起的 GRN 变化", options: { bold: true, breakLine: true, fontSize: 12 } },
|
| 448 |
+
{ text: "分别输入 ctrl / pert 表达", options: { breakLine: true, fontSize: 11, color: C.textS } },
|
| 449 |
+
{ text: "得到两个 attention matrix", options: { breakLine: true, fontSize: 11, color: C.textS } },
|
| 450 |
+
{ text: "差值 = 扰动引起的 GRN 变化", options: { breakLine: true, fontSize: 11, color: C.textS } },
|
| 451 |
+
], { x: 5.45, y: 2.1, w: 3.8, h: 2.1, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.25 });
|
| 452 |
+
// Formula highlight
|
| 453 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 454 |
+
x: 5.45, y: 4.3, w: 3.8, h: 0.45, fill: { color: C.tealLt },
|
| 455 |
+
});
|
| 456 |
+
s.addText("\u0394_attn = Attn(pert) - Attn(ctrl)", {
|
| 457 |
+
x: 5.45, y: 4.32, w: 3.8, h: 0.4,
|
| 458 |
+
fontSize: 14, fontFace: "Consolas", color: C.tealDk, bold: true, align: "center", margin: 0,
|
| 459 |
+
});
|
| 460 |
+
|
| 461 |
+
// ============================
|
| 462 |
+
addSlideNum(s);
|
| 463 |
+
// SLIDE 9: METHOD OVERVIEW
|
| 464 |
+
// ============================
|
| 465 |
+
s = pres.addSlide();
|
| 466 |
+
s.background = { color: C.light };
|
| 467 |
+
titleBar(s, "方法总览:Cascaded Flow Matching");
|
| 468 |
+
|
| 469 |
+
s.addText("在 scDFM 的 flow matching 框架上,引入 GRN-aware latent flow", {
|
| 470 |
+
x: 0.6, y: 1.05, w: 8.8, h: 0.3,
|
| 471 |
+
fontSize: 13, fontFace: "Calibri", color: C.textS, italic: true, margin: 0,
|
| 472 |
+
});
|
| 473 |
+
|
| 474 |
+
// Stage 1
|
| 475 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 476 |
+
x: 0.5, y: 1.55, w: 4.3, h: 2.3,
|
| 477 |
+
fill: { color: C.amber, transparency: 92 }, line: { color: C.amber, width: 2 },
|
| 478 |
+
});
|
| 479 |
+
s.addText("Stage 1: GRN Latent Flow", {
|
| 480 |
+
x: 0.7, y: 1.65, w: 3.9, h: 0.4,
|
| 481 |
+
fontSize: 17, fontFace: "Georgia", color: C.amber, bold: true, margin: 0,
|
| 482 |
+
});
|
| 483 |
+
s.addText([
|
| 484 |
+
{ text: "noise \u2192 GRN 变化特征", options: { bold: true, breakLine: true, fontSize: 14 } },
|
| 485 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 486 |
+
{ text: "理解调控网络如何改变", options: { breakLine: true, fontSize: 13, color: C.textS } },
|
| 487 |
+
{ text: "推理时先完成", options: { fontSize: 13, color: C.amber, bold: true } },
|
| 488 |
+
], { x: 0.7, y: 2.15, w: 3.9, h: 1.5, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.2 });
|
| 489 |
+
|
| 490 |
+
// Arrow
|
| 491 |
+
s.addText("\u2192", {
|
| 492 |
+
x: 4.5, y: 2.2, w: 1, h: 0.8,
|
| 493 |
+
fontSize: 40, fontFace: "Calibri", color: C.tealDk, align: "center", valign: "middle", margin: 0,
|
| 494 |
+
});
|
| 495 |
+
|
| 496 |
+
// Stage 2
|
| 497 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 498 |
+
x: 5.2, y: 1.55, w: 4.3, h: 2.3,
|
| 499 |
+
fill: { color: C.teal, transparency: 92 }, line: { color: C.teal, width: 2 },
|
| 500 |
+
});
|
| 501 |
+
s.addText("Stage 2: Expression Flow", {
|
| 502 |
+
x: 5.4, y: 1.65, w: 3.9, h: 0.4,
|
| 503 |
+
fontSize: 17, fontFace: "Georgia", color: C.teal, bold: true, margin: 0,
|
| 504 |
+
});
|
| 505 |
+
s.addText([
|
| 506 |
+
{ text: "noise \u2192 基因表达预测", options: { bold: true, breakLine: true, fontSize: 14 } },
|
| 507 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 508 |
+
{ text: "基于 GRN 变化预测表达", options: { breakLine: true, fontSize: 13, color: C.textS } },
|
| 509 |
+
{ text: "推理时后完成", options: { fontSize: 13, color: C.teal, bold: true } },
|
| 510 |
+
], { x: 5.4, y: 2.15, w: 3.9, h: 1.5, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.2 });
|
| 511 |
+
|
| 512 |
+
// Intuition banner
|
| 513 |
+
card(s, 0.5, 4.2, 9.0, 1.05, { accent: C.tealDk });
|
| 514 |
+
s.addText("生物学直觉", {
|
| 515 |
+
x: 0.75, y: 4.3, w: 2, h: 0.3,
|
| 516 |
+
fontSize: 14, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 517 |
+
});
|
| 518 |
+
s.addText("模型先\u201C想清楚\u201D扰动改变了哪些基因调控关系,再基于这些理解去预测表达变化", {
|
| 519 |
+
x: 0.75, y: 4.6, w: 8.5, h: 0.4,
|
| 520 |
+
fontSize: 13, fontFace: "Calibri", color: C.text, margin: 0,
|
| 521 |
+
});
|
| 522 |
+
|
| 523 |
+
// ============================
|
| 524 |
+
addSlideNum(s);
|
| 525 |
+
// SLIDE 10: ARCHITECTURE
|
| 526 |
+
// ============================
|
| 527 |
+
s = pres.addSlide();
|
| 528 |
+
s.background = { color: C.light };
|
| 529 |
+
titleBar(s, "模型架构");
|
| 530 |
+
|
| 531 |
+
// Left architecture diagram
|
| 532 |
+
// Condition inputs
|
| 533 |
+
card(s, 0.3, 1.15, 3.0, 1.35);
|
| 534 |
+
s.addText("条件信息(推理时可用)", {
|
| 535 |
+
x: 0.45, y: 1.2, w: 2.7, h: 0.25,
|
| 536 |
+
fontSize: 11, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 537 |
+
});
|
| 538 |
+
s.addText([
|
| 539 |
+
{ text: "x_ctrl (control 表达)", options: { breakLine: true } },
|
| 540 |
+
{ text: "pert_id (扰动基因)", options: { breakLine: true } },
|
| 541 |
+
{ text: "t\u2081, t\u2082 (时间步)", options: {} },
|
| 542 |
+
], { x: 0.45, y: 1.5, w: 2.7, h: 0.85, fontSize: 10.5, fontFace: "Consolas", color: C.text, margin: 0, bullet: true, paraSpaceAfter: 3 });
|
| 543 |
+
|
| 544 |
+
// GRN target
|
| 545 |
+
card(s, 3.6, 1.15, 3.1, 1.35, { fill: "FFF7ED" });
|
| 546 |
+
s.addText("辅助目标(从噪声生成)", {
|
| 547 |
+
x: 3.75, y: 1.2, w: 2.8, h: 0.25,
|
| 548 |
+
fontSize: 11, fontFace: "Calibri", color: C.amber, bold: true, margin: 0,
|
| 549 |
+
});
|
| 550 |
+
s.addText([
|
| 551 |
+
{ text: "z = \u0394_attn @ gene_emb", options: { breakLine: true, fontFace: "Consolas" } },
|
| 552 |
+
{ text: "GRN 变化特征 (512d/gene)", options: { breakLine: true } },
|
| 553 |
+
{ text: "来自 frozen scGPT", options: {} },
|
| 554 |
+
], { x: 3.75, y: 1.5, w: 2.8, h: 0.85, fontSize: 10.5, fontFace: "Calibri", color: C.text, margin: 0, bullet: true, paraSpaceAfter: 3 });
|
| 555 |
+
|
| 556 |
+
// Expression stream
|
| 557 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 558 |
+
x: 0.3, y: 2.75, w: 3.0, h: 0.5, fill: { color: C.teal, transparency: 85 }, line: { color: C.teal, width: 1 },
|
| 559 |
+
});
|
| 560 |
+
s.addText("Expression Stream \u2192 expr_tokens", {
|
| 561 |
+
x: 0.4, y: 2.78, w: 2.8, h: 0.45,
|
| 562 |
+
fontSize: 11, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0, valign: "middle",
|
| 563 |
+
});
|
| 564 |
+
|
| 565 |
+
// Latent stream
|
| 566 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 567 |
+
x: 3.6, y: 2.75, w: 3.1, h: 0.5, fill: { color: C.amber, transparency: 85 }, line: { color: C.amber, width: 1 },
|
| 568 |
+
});
|
| 569 |
+
s.addText("Latent Stream \u2192 lat_tokens", {
|
| 570 |
+
x: 3.7, y: 2.78, w: 2.9, h: 0.45,
|
| 571 |
+
fontSize: 11, fontFace: "Calibri", color: C.amber, bold: true, margin: 0, valign: "middle",
|
| 572 |
+
});
|
| 573 |
+
|
| 574 |
+
// Merge
|
| 575 |
+
s.addText("\u2295 加法融合", {
|
| 576 |
+
x: 1.5, y: 3.32, w: 3.5, h: 0.3,
|
| 577 |
+
fontSize: 12, fontFace: "Calibri", color: C.text, align: "center", margin: 0,
|
| 578 |
+
});
|
| 579 |
+
|
| 580 |
+
// Shared backbone
|
| 581 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 582 |
+
x: 0.3, y: 3.7, w: 6.4, h: 0.6, fill: { color: C.tealDk },
|
| 583 |
+
});
|
| 584 |
+
s.addText("Shared Backbone: DiffPerceiverBlock \u00D7 4 + GeneadaLN", {
|
| 585 |
+
x: 0.5, y: 3.73, w: 6.0, h: 0.55,
|
| 586 |
+
fontSize: 12.5, fontFace: "Calibri", color: C.white, bold: true, margin: 0, valign: "middle",
|
| 587 |
+
});
|
| 588 |
+
|
| 589 |
+
// Two heads
|
| 590 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.3, y: 4.5, w: 3.0, h: 0.45, fill: { color: C.teal } });
|
| 591 |
+
s.addText("ExprHead \u2192 v_expr (B,G)", {
|
| 592 |
+
x: 0.4, y: 4.52, w: 2.8, h: 0.4,
|
| 593 |
+
fontSize: 11, fontFace: "Calibri", color: C.white, bold: true, margin: 0, valign: "middle",
|
| 594 |
+
});
|
| 595 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 3.6, y: 4.5, w: 3.1, h: 0.45, fill: { color: C.amber } });
|
| 596 |
+
s.addText("LatentHead \u2192 v_latent (B,G,512)", {
|
| 597 |
+
x: 3.7, y: 4.52, w: 2.9, h: 0.4,
|
| 598 |
+
fontSize: 11, fontFace: "Calibri", color: C.white, bold: true, margin: 0, valign: "middle",
|
| 599 |
+
});
|
| 600 |
+
|
| 601 |
+
// Right: key design points
|
| 602 |
+
card(s, 7.0, 1.15, 2.7, 3.8);
|
| 603 |
+
s.addText("设计要点", {
|
| 604 |
+
x: 7.15, y: 1.25, w: 2.4, h: 0.3,
|
| 605 |
+
fontSize: 13, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 606 |
+
});
|
| 607 |
+
s.addText([
|
| 608 |
+
{ text: "双流输入", options: { bold: true, breakLine: true, fontSize: 11 } },
|
| 609 |
+
{ text: "表达 + GRN latent 各自编码后加法融合", options: { breakLine: true, fontSize: 10, color: C.textS } },
|
| 610 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 611 |
+
{ text: "共享骨干", options: { bold: true, breakLine: true, fontSize: 11 } },
|
| 612 |
+
{ text: "4 层 DiffPerceiverBlock 联合处理", options: { breakLine: true, fontSize: 10, color: C.textS } },
|
| 613 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 614 |
+
{ text: "双头输出", options: { bold: true, breakLine: true, fontSize: 11 } },
|
| 615 |
+
{ text: "分别预测表达和 latent 速度场", options: { breakLine: true, fontSize: 10, color: C.textS } },
|
| 616 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 617 |
+
{ text: "条件注入", options: { bold: true, breakLine: true, fontSize: 11 } },
|
| 618 |
+
{ text: "c = t\u2081 + t\u2082 + pert_emb", options: { breakLine: true, fontSize: 10, color: C.textS } },
|
| 619 |
+
{ text: "通过 adaLN 注入", options: { fontSize: 10, color: C.textS } },
|
| 620 |
+
], { x: 7.15, y: 1.6, w: 2.4, h: 3.2, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.15 });
|
| 621 |
+
|
| 622 |
+
// ============================
|
| 623 |
+
addSlideNum(s);
|
| 624 |
+
// SLIDE 11: TRAINING & INFERENCE
|
| 625 |
+
// ============================
|
| 626 |
+
s = pres.addSlide();
|
| 627 |
+
s.background = { color: C.light };
|
| 628 |
+
titleBar(s, "Cascaded 训练与推理");
|
| 629 |
+
|
| 630 |
+
// Training (left)
|
| 631 |
+
card(s, 0.5, 1.15, 4.3, 4.0, { accent: C.purple });
|
| 632 |
+
s.addText("训练:概率切换", {
|
| 633 |
+
x: 0.75, y: 1.25, w: 3.8, h: 0.35,
|
| 634 |
+
fontSize: 16, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 635 |
+
});
|
| 636 |
+
s.addText("不同时优化两个 flow,随机 coin flip:", {
|
| 637 |
+
x: 0.75, y: 1.65, w: 3.8, h: 0.25,
|
| 638 |
+
fontSize: 12, fontFace: "Calibri", color: C.textS, margin: 0,
|
| 639 |
+
});
|
| 640 |
+
|
| 641 |
+
// 40% latent
|
| 642 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 643 |
+
x: 0.75, y: 2.1, w: 3.8, h: 0.9,
|
| 644 |
+
fill: { color: C.amber, transparency: 90 }, line: { color: C.amber, width: 1 },
|
| 645 |
+
});
|
| 646 |
+
s.addText("40%", {
|
| 647 |
+
x: 0.85, y: 2.15, w: 0.8, h: 0.35,
|
| 648 |
+
fontSize: 22, fontFace: "Georgia", color: C.amber, bold: true, margin: 0,
|
| 649 |
+
});
|
| 650 |
+
s.addText([
|
| 651 |
+
{ text: "训练 Latent Flow", options: { bold: true, breakLine: true } },
|
| 652 |
+
{ text: "t\u2082 随机, t\u2081=0, 只算 loss_latent", options: {} },
|
| 653 |
+
], { x: 1.7, y: 2.15, w: 2.7, h: 0.75, fontSize: 11, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.5 });
|
| 654 |
+
|
| 655 |
+
// 60% expression
|
| 656 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 657 |
+
x: 0.75, y: 3.2, w: 3.8, h: 0.9,
|
| 658 |
+
fill: { color: C.teal, transparency: 90 }, line: { color: C.teal, width: 1 },
|
| 659 |
+
});
|
| 660 |
+
s.addText("60%", {
|
| 661 |
+
x: 0.85, y: 3.25, w: 0.8, h: 0.35,
|
| 662 |
+
fontSize: 22, fontFace: "Georgia", color: C.teal, bold: true, margin: 0,
|
| 663 |
+
});
|
| 664 |
+
s.addText([
|
| 665 |
+
{ text: "训练 Expression Flow", options: { bold: true, breakLine: true } },
|
| 666 |
+
{ text: "t\u2081 随机, t\u2082\u22481, 只算 loss_expr", options: {} },
|
| 667 |
+
], { x: 1.7, y: 3.25, w: 2.7, h: 0.75, fontSize: 11, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.5 });
|
| 668 |
+
|
| 669 |
+
// Inference (right)
|
| 670 |
+
card(s, 5.2, 1.15, 4.3, 4.0, { accent: C.teal });
|
| 671 |
+
s.addText("推理:两阶段串行", {
|
| 672 |
+
x: 5.45, y: 1.25, w: 3.8, h: 0.35,
|
| 673 |
+
fontSize: 16, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 674 |
+
});
|
| 675 |
+
|
| 676 |
+
// Stage 1
|
| 677 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 678 |
+
x: 5.45, y: 1.8, w: 3.8, h: 1.2,
|
| 679 |
+
fill: { color: C.amber, transparency: 90 }, line: { color: C.amber, width: 1 },
|
| 680 |
+
});
|
| 681 |
+
s.addText("Stage 1: GRN Latent", {
|
| 682 |
+
x: 5.55, y: 1.85, w: 3.6, h: 0.3,
|
| 683 |
+
fontSize: 13, fontFace: "Calibri", color: C.amber, bold: true, margin: 0,
|
| 684 |
+
});
|
| 685 |
+
s.addText([
|
| 686 |
+
{ text: "z_noise ==(ODE)==> z_clean", options: { breakLine: true, fontFace: "Consolas", fontSize: 11 } },
|
| 687 |
+
{ text: "先理解 GRN 如何变化 (t\u2082: 0\u21921)", options: { fontSize: 11 } },
|
| 688 |
+
], { x: 5.55, y: 2.2, w: 3.6, h: 0.65, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.4 });
|
| 689 |
+
|
| 690 |
+
// Arrow
|
| 691 |
+
s.addText("\u2193", {
|
| 692 |
+
x: 5.45, y: 3.05, w: 3.8, h: 0.35,
|
| 693 |
+
fontSize: 22, fontFace: "Calibri", color: C.tealDk, align: "center", margin: 0,
|
| 694 |
+
});
|
| 695 |
+
|
| 696 |
+
// Stage 2
|
| 697 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 698 |
+
x: 5.45, y: 3.45, w: 3.8, h: 1.2,
|
| 699 |
+
fill: { color: C.teal, transparency: 90 }, line: { color: C.teal, width: 1 },
|
| 700 |
+
});
|
| 701 |
+
s.addText("Stage 2: Expression", {
|
| 702 |
+
x: 5.55, y: 3.5, w: 3.6, h: 0.3,
|
| 703 |
+
fontSize: 13, fontFace: "Calibri", color: C.teal, bold: true, margin: 0,
|
| 704 |
+
});
|
| 705 |
+
s.addText([
|
| 706 |
+
{ text: "x_noise ==(ODE)==> x_pred", options: { breakLine: true, fontFace: "Consolas", fontSize: 11 } },
|
| 707 |
+
{ text: "基于 z_clean 预测表达 (t\u2081: 0\u21921)", options: { fontSize: 11 } },
|
| 708 |
+
], { x: 5.55, y: 3.85, w: 3.6, h: 0.65, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.4 });
|
| 709 |
+
|
| 710 |
+
// ============================
|
| 711 |
+
addSlideNum(s);
|
| 712 |
+
// SLIDE 12: CHALLENGES
|
| 713 |
+
// ============================
|
| 714 |
+
s = pres.addSlide();
|
| 715 |
+
s.background = { color: C.light };
|
| 716 |
+
titleBar(s, "当前挑战与解决方向");
|
| 717 |
+
|
| 718 |
+
// Challenge 1
|
| 719 |
+
card(s, 0.5, 1.15, 4.3, 4.0, { accent: C.red });
|
| 720 |
+
s.addText("挑战 1:GRN 信号噪声大", {
|
| 721 |
+
x: 0.75, y: 1.25, w: 3.8, h: 0.3,
|
| 722 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 723 |
+
});
|
| 724 |
+
s.addText([
|
| 725 |
+
{ text: "Attention: 5000\u00D75000 = 25M 非零值", options: { breakLine: true } },
|
| 726 |
+
{ text: "真实 GRN: 每基因仅 ~20-50 靶标", options: { breakLine: true } },
|
| 727 |
+
{ text: "99%+ attention 值是噪声", options: { breakLine: true, color: C.red, bold: true } },
|
| 728 |
+
{ text: "latent loss \u2248 1.12 >> expr loss \u2248 0.019", options: { color: C.red } },
|
| 729 |
+
], { x: 0.75, y: 1.65, w: 3.8, h: 1.15, fontSize: 12, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 730 |
+
|
| 731 |
+
// Solution 1
|
| 732 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 733 |
+
x: 0.75, y: 3.05, w: 3.8, h: 1.8,
|
| 734 |
+
fill: { color: C.green, transparency: 90 }, line: { color: C.green, width: 1 },
|
| 735 |
+
});
|
| 736 |
+
s.addText("解决:稀疏化 Top-K", {
|
| 737 |
+
x: 0.85, y: 3.1, w: 3.6, h: 0.3,
|
| 738 |
+
fontSize: 13, fontFace: "Calibri", color: C.green, bold: true, margin: 0,
|
| 739 |
+
});
|
| 740 |
+
s.addText([
|
| 741 |
+
{ text: "每个基因只保留 |\u0394| 最大的 K=30 个", options: { breakLine: true } },
|
| 742 |
+
{ text: "过滤 99.4% 噪声", options: { breakLine: true, bold: true } },
|
| 743 |
+
{ text: "sparse_topk_emb 模式", options: { fontFace: "Consolas" } },
|
| 744 |
+
], { x: 0.85, y: 3.45, w: 3.6, h: 1.0, fontSize: 11.5, fontFace: "Calibri", color: C.text, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 745 |
+
|
| 746 |
+
// Challenge 2
|
| 747 |
+
card(s, 5.2, 1.15, 4.3, 4.0, { accent: C.red });
|
| 748 |
+
s.addText("挑战 2:512 维 Latent 太难预测", {
|
| 749 |
+
x: 5.45, y: 1.25, w: 3.8, h: 0.3,
|
| 750 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, bold: true, margin: 0,
|
| 751 |
+
});
|
| 752 |
+
s.addText([
|
| 753 |
+
{ text: "每基因 512 维 = 250 万维速度场", options: { breakLine: true } },
|
| 754 |
+
{ text: "模型每步需预测如此大的向量", options: { breakLine: true } },
|
| 755 |
+
{ text: "消融: 512\u21921 维, loss 从 ~1.1 降到 ~0.5", options: { breakLine: true, color: C.red, bold: true } },
|
| 756 |
+
{ text: "维度是难度的重要来源", options: { color: C.red } },
|
| 757 |
+
], { x: 5.45, y: 1.65, w: 3.8, h: 1.15, fontSize: 12, fontFace: "Calibri", color: C.textS, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 758 |
+
|
| 759 |
+
// Solution 2
|
| 760 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 761 |
+
x: 5.45, y: 3.05, w: 3.8, h: 1.8,
|
| 762 |
+
fill: { color: C.green, transparency: 90 }, line: { color: C.green, width: 1 },
|
| 763 |
+
});
|
| 764 |
+
s.addText("解决:PCA 降维", {
|
| 765 |
+
x: 5.55, y: 3.1, w: 3.6, h: 0.3,
|
| 766 |
+
fontSize: 13, fontFace: "Calibri", color: C.green, bold: true, margin: 0,
|
| 767 |
+
});
|
| 768 |
+
s.addText([
|
| 769 |
+
{ text: "512-d gene_emb \u2192 PCA 投影到 64 维", options: { breakLine: true } },
|
| 770 |
+
{ text: "去掉冗余维度,保留主变化方向", options: { breakLine: true, bold: true } },
|
| 771 |
+
{ text: "sparse_pca 模式", options: { fontFace: "Consolas" } },
|
| 772 |
+
], { x: 5.55, y: 3.45, w: 3.6, h: 1.0, fontSize: 11.5, fontFace: "Calibri", color: C.text, margin: 0, bullet: true, paraSpaceAfter: 4 });
|
| 773 |
+
|
| 774 |
+
// ============================
|
| 775 |
+
addSlideNum(s);
|
| 776 |
+
// SLIDE 13: SUMMARY & FUTURE
|
| 777 |
+
// ============================
|
| 778 |
+
s = pres.addSlide();
|
| 779 |
+
s.background = { color: C.light };
|
| 780 |
+
titleBar(s, "总结与展望");
|
| 781 |
+
|
| 782 |
+
// Core contribution
|
| 783 |
+
card(s, 0.5, 1.15, 9.0, 1.4, { accent: C.teal });
|
| 784 |
+
s.addText("核心贡献", {
|
| 785 |
+
x: 0.75, y: 1.25, w: 2, h: 0.3,
|
| 786 |
+
fontSize: 16, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 787 |
+
});
|
| 788 |
+
s.addText("不是架构改进,而是从生物学机制出发重新建模:用 Cascaded Flow Matching 实现\u201C先理解调控变化,再预测表达变化\u201D", {
|
| 789 |
+
x: 0.75, y: 1.6, w: 8.5, h: 0.75,
|
| 790 |
+
fontSize: 14, fontFace: "Calibri", color: C.text, margin: 0, lineSpacingMultiple: 1.35,
|
| 791 |
+
});
|
| 792 |
+
|
| 793 |
+
// Future experiment
|
| 794 |
+
s.addText("后续关键实验:验证因果假��", {
|
| 795 |
+
x: 0.5, y: 2.8, w: 5, h: 0.35,
|
| 796 |
+
fontSize: 16, fontFace: "Calibri", color: C.tealDk, bold: true, margin: 0,
|
| 797 |
+
});
|
| 798 |
+
|
| 799 |
+
const tbl = [
|
| 800 |
+
[
|
| 801 |
+
{ text: "推理方式", options: { bold: true, color: "FFFFFF", fill: { color: C.tealDk } } },
|
| 802 |
+
{ text: "含义", options: { bold: true, color: "FFFFFF", fill: { color: C.tealDk } } },
|
| 803 |
+
{ text: "预期", options: { bold: true, color: "FFFFFF", fill: { color: C.tealDk } } },
|
| 804 |
+
],
|
| 805 |
+
[
|
| 806 |
+
{ text: "先 GRN \u2192 后 Expression", options: { bold: true, fill: { color: C.tealLt } } },
|
| 807 |
+
{ text: "先理解调控变化,再预测表达", options: { fill: { color: C.tealLt } } },
|
| 808 |
+
{ text: "最优", options: { bold: true, color: C.teal, fill: { color: C.tealLt } } },
|
| 809 |
+
],
|
| 810 |
+
["先 Expression \u2192 后 GRN", "先预测表达,再理解调控", "次优"],
|
| 811 |
+
["同时 random", "无显式顺序", "最差"],
|
| 812 |
+
];
|
| 813 |
+
s.addTable(tbl, {
|
| 814 |
+
x: 0.5, y: 3.25, w: 9.0,
|
| 815 |
+
fontSize: 12, fontFace: "Calibri",
|
| 816 |
+
border: { pt: 0.5, color: C.border },
|
| 817 |
+
colW: [3.0, 3.5, 2.5],
|
| 818 |
+
rowH: [0.38, 0.38, 0.38, 0.38],
|
| 819 |
+
autoPage: false,
|
| 820 |
+
});
|
| 821 |
+
|
| 822 |
+
s.addText("如果\u201C先 GRN 后 Expression\u201D显著优于其他 \u2192 验证 GRN 理解是预测表达变化的前提", {
|
| 823 |
+
x: 0.5, y: 4.85, w: 9.0, h: 0.35,
|
| 824 |
+
fontSize: 12, fontFace: "Calibri", color: C.teal, italic: true, bold: true, margin: 0,
|
| 825 |
+
});
|
| 826 |
+
|
| 827 |
+
// ============================
|
| 828 |
+
addSlideNum(s);
|
| 829 |
+
// SLIDE 14: CONCLUSION
|
| 830 |
+
// ============================
|
| 831 |
+
s = pres.addSlide();
|
| 832 |
+
s.background = { color: C.dark };
|
| 833 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0, y: 0, w: 0.15, h: 5.625, fill: { color: C.teal } });
|
| 834 |
+
|
| 835 |
+
s.addText("核心结论", {
|
| 836 |
+
x: 0.8, y: 1.2, w: 3, h: 0.4,
|
| 837 |
+
fontSize: 16, fontFace: "Calibri", color: "CBD5E1", margin: 0,
|
| 838 |
+
});
|
| 839 |
+
s.addText("用 scGPT 的 attention delta 显式提取扰动引起的基因调控网络变化,通过 cascaded flow matching 强制模型\u201C先理解 GRN 如何改变,再预测表达如何变化\u201D,从而将生物学先验融入生成式模型的推理过程。", {
|
| 840 |
+
x: 0.8, y: 1.8, w: 8.4, h: 2.2,
|
| 841 |
+
fontSize: 22, fontFace: "Georgia", color: C.white, lineSpacingMultiple: 1.45, margin: 0,
|
| 842 |
+
});
|
| 843 |
+
s.addText("GRN-Guided Cascaded Flow Matching", {
|
| 844 |
+
x: 0.8, y: 4.6, w: 5, h: 0.35,
|
| 845 |
+
fontSize: 14, fontFace: "Calibri", color: "22D3EE", margin: 0,
|
| 846 |
+
});
|
| 847 |
+
|
| 848 |
+
addSlideNum(s, true);
|
| 849 |
+
|
| 850 |
+
// === SAVE ===
|
| 851 |
+
const outPath = "/home/hp250092/ku50001222/qian/aivc/lfj/Report/PPT/GRN_CCFM_presentation.pptx";
|
| 852 |
+
pres.writeFile({ fileName: outPath })
|
| 853 |
+
.then(() => console.log("Saved: " + outPath))
|
| 854 |
+
.catch(err => console.error("Error:", err));
|
Report/PPT/fixed-01.jpg
ADDED
|
Git LFS Details
|
Report/PPT/fixed14-14.jpg
ADDED
|
Git LFS Details
|
Report/PPT/fixed6-06.jpg
ADDED
|
Git LFS Details
|
Report/PPT/fixed8-08.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-01.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-02.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-03.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-04.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-05.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-06.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-07.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-08.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-09.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-10.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-11.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-12.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-13.jpg
ADDED
|
Git LFS Details
|
Report/PPT/slide-14.jpg
ADDED
|
Git LFS Details
|
Report/PPT/讲解稿.md
ADDED
|
@@ -0,0 +1,183 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GRN-Guided Cascaded Flow Matching 讲解稿
|
| 2 |
+
|
| 3 |
+
---
|
| 4 |
+
|
| 5 |
+
## Slide 1:封面
|
| 6 |
+
|
| 7 |
+
大家好,今天我汇报的题目是 **GRN-Guided Cascaded Flow Matching for Single-Cell Perturbation Prediction**——用基因调控网络引导的级联流匹配方法来做单细胞扰动预测。
|
| 8 |
+
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
## Slide 2:Task——单细胞扰动预测
|
| 12 |
+
|
| 13 |
+
首先介绍一下我们要解决的任务。
|
| 14 |
+
|
| 15 |
+
**虚拟细胞**是当前计算生物学的核心愿景:我们希望构建一个 AI 模型,能够在计算机中模拟真实细胞的行为——给定任意输入条件,预测细胞的分子状态变化。单细胞扰动预测是实现虚拟细胞最关键的子任务之一。
|
| 16 |
+
|
| 17 |
+
扰动有很多种类型:药物扰动、细胞因子扰动、基因扰动等。我们这个工作聚焦的是**基因扰动**,具体来说就是用 CRISPR 技术对细胞进行基因敲除或过表达,然后用单细胞 RNA-seq 测量所有基因的表达变化。
|
| 18 |
+
|
| 19 |
+
任务的形式化很简单:已知 control 细胞的基因表达谱和被扰动的基因,预测扰动后的基因表达谱。基因数量大约是 5000 个高变基因。
|
| 20 |
+
|
| 21 |
+
---
|
| 22 |
+
|
| 23 |
+
## Slide 3:为什么重要 & 数据特点
|
| 24 |
+
|
| 25 |
+
这个任务为什么重要?三个原因:
|
| 26 |
+
|
| 27 |
+
第一,**药物筛选加速**。做一次 Perturb-seq 实验成本非常高,如果能用计算预测,可以大幅缩小候选范围。
|
| 28 |
+
|
| 29 |
+
第二,**组合扰动爆炸**。N 个基因的两两组合就是 N(N-1)/2 种实验,不可能穷举,必须靠预测。
|
| 30 |
+
|
| 31 |
+
第三,**理解疾病机制**。预测哪些基因被扰动后会产生某种疾病表型。
|
| 32 |
+
|
| 33 |
+
我们使用的是 Norman 数据集,大约 9000 个细胞,5000 个高变基因,105 种单基因和双基因的 CRISPR 扰动。
|
| 34 |
+
|
| 35 |
+
这里有一个关键挑战:**细胞配对不可得**。扰动是破坏性的,一个细胞只能测一次,我们无法得到同一个细胞扰动前后的配对数据。
|
| 36 |
+
|
| 37 |
+
---
|
| 38 |
+
|
| 39 |
+
## Slide 4:现有方法——简单基线与预训练大模型
|
| 40 |
+
|
| 41 |
+
接下来看看现有的方法。
|
| 42 |
+
|
| 43 |
+
最简单的基线是 **Additive Shift**,也就是均值偏移:直接用训练集里扰动前后的平均差来预测。它假设扰动效应对所有细胞是一个常数平移,完全忽略了细胞异质性。但有意思的是,这个简单基线出奇地难以超越——很多复杂模型在 top DE 基因上并不比它好。
|
| 44 |
+
|
| 45 |
+
**scGPT** 是 2024 年发表在 Nature Methods 上的工作,用自回归 Transformer 在大规模单细胞数据上预训练。它做扰动预测的方式是把扰动基因 mask 掉让模型补全。但本质上它是自回归补全,不是为扰动预测专门设计的。
|
| 46 |
+
|
| 47 |
+
**Geneformer** 也是 2024 年发在 Nature 上,用 rank-value encoding 做预训练。它做扰动预测是直接删掉目标基因的 token,看 embedding 变化。这是一个启发式方法,没有真正学习扰动的动力学。
|
| 48 |
+
|
| 49 |
+
**CPA** 把细胞状态分解为 basal state 加 perturbation effect,在 latent space 里做线性组合。问题是线性可加假设太强了,基因调控本质上是非线性的。
|
| 50 |
+
|
| 51 |
+
---
|
| 52 |
+
|
| 53 |
+
## Slide 5:现有方法——专用扰动预测模型
|
| 54 |
+
|
| 55 |
+
再看看专门为扰动预测设计的模型。
|
| 56 |
+
|
| 57 |
+
**GEARS** 用 Gene Ontology 图上的 GNN 来编码基因关系,但 GO 图是静态的先验知识,不随细胞状态变化,而且它是确定性预测,不能给出分布。
|
| 58 |
+
|
| 59 |
+
**STATE** 是 ICLR 2025 的工作,用 Stacked Attention 做表达变换,同样是确定性预测,没有从 GRN 变化的角度建模。
|
| 60 |
+
|
| 61 |
+
**CellFlow** 也用了 flow matching 框架,但它用预训练 embedding 作为条件,这些 embedding 编码的是绝对状态,没有显式建模扰动对调控网络的改变。
|
| 62 |
+
|
| 63 |
+
**scDFM** 是我们的基线方法,今年发表在 ICLR 2026。它把 Conditional Flow Matching 引入扰动预测,学习从噪声到目标表达的速度场。优点是生成式模型,能给出分布,训练也稳定。但问题在于信息来源单一——只有 control 的数值表达加扰动基因的 embedding,模型不理解基因间的调控关系。
|
| 64 |
+
|
| 65 |
+
---
|
| 66 |
+
|
| 67 |
+
## Slide 6:所有现有方法的共同盲区
|
| 68 |
+
|
| 69 |
+
总结一下,所有现有方法都有一个共同的盲区:
|
| 70 |
+
|
| 71 |
+
它们都是 **扰动 → 黑箱模型 → 表达变化** 这样的端到端模式。
|
| 72 |
+
|
| 73 |
+
没有任何一个方法显式地建模中间这一步:**扰动是如何通过基因调控网络的变化来导致表达变化的**。
|
| 74 |
+
|
| 75 |
+
我们的方法要做的,就是把这一步补上:**扰动 → GRN 变化 → 表达变化**,显式建模生物学机制。
|
| 76 |
+
|
| 77 |
+
---
|
| 78 |
+
|
| 79 |
+
## Slide 7:Motivation 1——Flow Matching 解决配对问题
|
| 80 |
+
|
| 81 |
+
我们的工作有三个 motivation。
|
| 82 |
+
|
| 83 |
+
第一个是用 Flow Matching 来解决细胞配对问题。刚才提到,单细胞扰动数据天然没有 paired data,一个细胞扰动后就变了,无法回到扰动前状态。
|
| 84 |
+
|
| 85 |
+
传统方法要么用群体均值匹配丢失异质性,要么用 Autoencoder 受限于重建质量。
|
| 86 |
+
|
| 87 |
+
**Flow Matching 的优势**在于:它学习的是从 source 分布到 target 分布的概率传输映���,天然适合 unpaired 数据。不需要逐细胞配对,只需要两组细胞的群体分布。通过 Conditional Optimal Transport 构造训练对,效率更高。而且它是生成式输出,每个 control 细胞可以采样多个预测,给出不确定性估计。
|
| 88 |
+
|
| 89 |
+
---
|
| 90 |
+
|
| 91 |
+
## Slide 8:Motivation 2 & 3——GRN 视角 + scGPT Attention
|
| 92 |
+
|
| 93 |
+
第二个 motivation 是从 GRN 变化的角度来理解扰动。
|
| 94 |
+
|
| 95 |
+
生物学上,基因扰动不是简单地改变一个基因的值。比如 CRISPR 敲除基因 A,首先 A 的表达降为 0,然后 A 直接调控的基因 B、C、D 发生变化,再往下 B 调控的 E、F,C 调控的 G、H 依次改变——这是一个通过基因调控网络的级联传播过程。如果我们能先理解 GRN 如何变化,再预测表达,预测会更准确。
|
| 96 |
+
|
| 97 |
+
第三个 motivation 是 scGPT 的 attention matrix 可以作为数据驱动的 GRN。预训练的 scGPT 在 attention matrix 里编码了基因间的调控关系:attn[i][j] 高表示基因 j 对基因 i 有强调控。而且这个 GRN 是上下文相关的,随细胞状态变化,比静态的 GO 图灵活得多。
|
| 98 |
+
|
| 99 |
+
我们可以分别输入 control 和 perturbed 的表达,得到两个 attention matrix,它们的差值 Δ_attn 就直接反映了扰动引起的 GRN 变化。
|
| 100 |
+
|
| 101 |
+
---
|
| 102 |
+
|
| 103 |
+
## Slide 9:方法总览——Cascaded Flow Matching
|
| 104 |
+
|
| 105 |
+
基于这三个 motivation,我们提出了 Cascaded Flow Matching。
|
| 106 |
+
|
| 107 |
+
核心思路是在 scDFM 的 flow matching 框架上,引入一个 **GRN-aware latent flow**,形成两阶段的级联结构:
|
| 108 |
+
|
| 109 |
+
**Stage 1 是 GRN Latent Flow**:从噪声生成 GRN 变化特征,理解调控网络如何改变。推理时先完成。
|
| 110 |
+
|
| 111 |
+
**Stage 2 是 Expression Flow**:从噪声生成基因表达预测,基于 GRN 变化来预测表达。推理时后完成。
|
| 112 |
+
|
| 113 |
+
生物学直觉很简单:模型先"想清楚"扰动改变了哪些基因调控关系,再基于这些理解去预测表达变化。
|
| 114 |
+
|
| 115 |
+
---
|
| 116 |
+
|
| 117 |
+
## Slide 10:模型架构
|
| 118 |
+
|
| 119 |
+
具体的模型架构是这样的。
|
| 120 |
+
|
| 121 |
+
输入分为两部分:一是条件信息,包括 control 表达、扰动基因 ID 和两个时间步;二是辅助生成目标,就是 GRN 变化特征 z,通过 frozen scGPT 的 Δ_attn 乘以 gene embedding 得到。
|
| 122 |
+
|
| 123 |
+
模型有**双流输入**:Expression Stream 编码表达信息,Latent Stream 编码 GRN latent 信息,两者加法融合。
|
| 124 |
+
|
| 125 |
+
融合后进入**共享骨干**:4 层 DiffPerceiverBlock,配合 GeneadaLN 注入条件信息。条件向量 c 由两个时间步加扰动 embedding 组成。
|
| 126 |
+
|
| 127 |
+
最后是**双头输出**:ExprHead 预测表达速度场,LatentHead 预测 latent 速度场。
|
| 128 |
+
|
| 129 |
+
---
|
| 130 |
+
|
| 131 |
+
## Slide 11:Cascaded 训练与推理
|
| 132 |
+
|
| 133 |
+
训练的时候我们不同时优化两个 flow,而是概率切换:
|
| 134 |
+
|
| 135 |
+
40% 的概率训练 Latent Flow——t₂ 随机采样,t₁ 固定为 0,只算 latent loss。
|
| 136 |
+
|
| 137 |
+
60% 的概率训练 Expression Flow——t₁ 随机采样,t₂ 接近 1,只算 expression loss。
|
| 138 |
+
|
| 139 |
+
推理的时候是两阶段串行:
|
| 140 |
+
|
| 141 |
+
**Stage 1** 先跑 GRN Latent,从 z_noise 通过 ODE 生成 z_clean,理解 GRN 如何变化。
|
| 142 |
+
|
| 143 |
+
**Stage 2** 再跑 Expression,利用已完成的 z_clean 作为条件,从 x_noise 通过 ODE 生成 x_pred。
|
| 144 |
+
|
| 145 |
+
这种先后顺序的设计,就是我们这个工作的核心:先理解调控变化,再预测表达变化。
|
| 146 |
+
|
| 147 |
+
---
|
| 148 |
+
|
| 149 |
+
## Slide 12:当前挑战与解决方向
|
| 150 |
+
|
| 151 |
+
目前有两个主要挑战。
|
| 152 |
+
|
| 153 |
+
**第一个挑战:GRN 信号噪声太大。** scGPT 的 attention matrix 是 5000×5000 的稠密矩阵,有 2500 万个非零值。但真实的 GRN 是极度稀疏的,一个基因通常只直接调控几十个靶标。所以 99% 以上的 attention 值都是噪声。实验也验证了这一点:latent loss 约 1.12,远高于 expression loss 的 0.019。
|
| 154 |
+
|
| 155 |
+
我们的解决方案是**稀疏化 Top-K**:每个基因只保留 Δ 值最大的 K=30 个连接,过滤掉 99.4% 的噪声。
|
| 156 |
+
|
| 157 |
+
**第二个挑战:512 维的 latent 太难预测。** 每个基因的 GRN 特征是 512 维,整个速度场是 250 万维,模型难以在每个时间步预测这么大的向量。消融实验证实,把维度从 512 降到 1,loss 从约 1.1 降到约 0.5。
|
| 158 |
+
|
| 159 |
+
解决方案是 **PCA 降维**:把 512 维的 gene embedding 通过 PCA 投影到 64 维,去掉冗余维度,只保留主要的变化方向。
|
| 160 |
+
|
| 161 |
+
---
|
| 162 |
+
|
| 163 |
+
## Slide 13:总结与展望
|
| 164 |
+
|
| 165 |
+
总结一下,我们这个工作的核心贡献不是在模型架构上做改进,而是**从生物学机制出发**重新建模扰动预测任务:用 Cascaded Flow Matching 实现"先理解调控变化,再预测表达变化"。
|
| 166 |
+
|
| 167 |
+
后续最关键的实验是验证因果假设。我们计划训练一个支持任意推理顺序的模型,然后对比三种推理方式:
|
| 168 |
+
|
| 169 |
+
- 先 GRN 后 Expression——我们预期这是最优的;
|
| 170 |
+
- 先 Expression 后 GRN——预期次优;
|
| 171 |
+
- 同时 random——预期最差。
|
| 172 |
+
|
| 173 |
+
如果"先 GRN 后 Expression"显著优于其他顺序,就验证了我们的核心假设:**理解基因调控网络的变化,是预测扰动表达变化的前提条件,而不是副产物。**
|
| 174 |
+
|
| 175 |
+
---
|
| 176 |
+
|
| 177 |
+
## Slide 14:核心结论
|
| 178 |
+
|
| 179 |
+
最后一句话总结:
|
| 180 |
+
|
| 181 |
+
我们用 scGPT 的 attention delta 显式提取扰动引起的基因调控网络变化,通过 cascaded flow matching 强制模型"先理解 GRN 如何改变,再预测表达如何变化",从而将生物学先验——扰动通过 GRN 级联传播——融入生成式模型的推理过程。
|
| 182 |
+
|
| 183 |
+
谢谢大家,欢迎提问。
|
Report/PPT2/GRN_CCFM_presentation.pdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2a364adbc69c18b72d658c56eab80dc2f46a3a7ee8e04d82a5f4524ec453efe3
|
| 3 |
+
size 199553
|
Report/PPT2/GRN_CCFM_presentation.pptx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0a1aad9d0018851617a8642f4a3afc393bfc330232ea929f581683af8b53ce08
|
| 3 |
+
size 317483
|
Report/PPT2/generate.js
ADDED
|
@@ -0,0 +1,781 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
const pptxgen = require("pptxgenjs");
|
| 2 |
+
const pres = new pptxgen();
|
| 3 |
+
pres.layout = "LAYOUT_16x9"; // 10" x 5.625"
|
| 4 |
+
pres.author = "Qian";
|
| 5 |
+
pres.title = "GRN-Guided Cascaded Flow Matching for Single-Cell Perturbation Prediction";
|
| 6 |
+
|
| 7 |
+
// ─── COLOR PALETTE ───
|
| 8 |
+
const NAVY = "1B2A4A";
|
| 9 |
+
const BLUE = "2E6B9E";
|
| 10 |
+
const LT_BLUE = "B8D4E8";
|
| 11 |
+
const BG_GRAY = "F0F2F5";
|
| 12 |
+
const CARD_BG = "F7F8FA";
|
| 13 |
+
const TXT = "2C3E50";
|
| 14 |
+
const TXT_MID = "4A5568";
|
| 15 |
+
const TXT_LT = "6B7B8D";
|
| 16 |
+
const WHITE = "FFFFFF";
|
| 17 |
+
const RED = "C0392B";
|
| 18 |
+
const GREEN = "27AE60";
|
| 19 |
+
const ORANGE = "D35400";
|
| 20 |
+
|
| 21 |
+
const HF = "Cambria";
|
| 22 |
+
const BF = "Calibri";
|
| 23 |
+
const CF = "Consolas";
|
| 24 |
+
|
| 25 |
+
// ─── HELPERS ───
|
| 26 |
+
function slideNum(slide, n) {
|
| 27 |
+
slide.addText(String(n), {
|
| 28 |
+
x: 9.2, y: 5.2, w: 0.5, h: 0.3,
|
| 29 |
+
fontSize: 10, color: TXT_LT, fontFace: BF, align: "right"
|
| 30 |
+
});
|
| 31 |
+
}
|
| 32 |
+
|
| 33 |
+
function headerBar(slide, title) {
|
| 34 |
+
slide.addShape(pres.shapes.RECTANGLE, {
|
| 35 |
+
x: 0, y: 0, w: 10, h: 0.85,
|
| 36 |
+
fill: { color: NAVY }
|
| 37 |
+
});
|
| 38 |
+
slide.addText(title, {
|
| 39 |
+
x: 0.6, y: 0.12, w: 8.8, h: 0.6,
|
| 40 |
+
fontSize: 24, fontFace: HF, color: WHITE, bold: true, margin: 0
|
| 41 |
+
});
|
| 42 |
+
}
|
| 43 |
+
|
| 44 |
+
function sectionLabel(slide, text, x, y, w) {
|
| 45 |
+
slide.addText(text, {
|
| 46 |
+
x: x || 0.6, y: y || 1.1, w: w || 8.8, h: 0.4,
|
| 47 |
+
fontSize: 17, fontFace: HF, color: NAVY, bold: true, margin: 0
|
| 48 |
+
});
|
| 49 |
+
}
|
| 50 |
+
|
| 51 |
+
function card(slide, x, y, w, h, accentColor) {
|
| 52 |
+
slide.addShape(pres.shapes.RECTANGLE, {
|
| 53 |
+
x, y, w, h, fill: { color: CARD_BG }
|
| 54 |
+
});
|
| 55 |
+
if (accentColor) {
|
| 56 |
+
slide.addShape(pres.shapes.RECTANGLE, {
|
| 57 |
+
x, y, w: 0.06, h, fill: { color: accentColor }
|
| 58 |
+
});
|
| 59 |
+
}
|
| 60 |
+
}
|
| 61 |
+
|
| 62 |
+
// ════════════════════════════════════════════════════════════
|
| 63 |
+
// SLIDE 1: TITLE
|
| 64 |
+
// ════════════════════════════════════════════════════════════
|
| 65 |
+
let s1 = pres.addSlide();
|
| 66 |
+
s1.background = { color: NAVY };
|
| 67 |
+
s1.addShape(pres.shapes.RECTANGLE, {
|
| 68 |
+
x: 0, y: 0, w: 10, h: 0.06, fill: { color: BLUE }
|
| 69 |
+
});
|
| 70 |
+
s1.addText([
|
| 71 |
+
{ text: "GRN-Guided Cascaded Flow Matching", options: { breakLine: true, fontSize: 34 } },
|
| 72 |
+
{ text: "for Single-Cell Perturbation Prediction", options: { fontSize: 28 } }
|
| 73 |
+
], {
|
| 74 |
+
x: 0.8, y: 1.0, w: 8.4, h: 2.2,
|
| 75 |
+
fontFace: HF, color: WHITE, bold: true,
|
| 76 |
+
align: "center", valign: "middle", paraSpaceAfter: 8
|
| 77 |
+
});
|
| 78 |
+
s1.addShape(pres.shapes.RECTANGLE, {
|
| 79 |
+
x: 3.8, y: 3.4, w: 2.4, h: 0.035, fill: { color: BLUE }
|
| 80 |
+
});
|
| 81 |
+
s1.addText("Group Meeting Report", {
|
| 82 |
+
x: 1, y: 3.65, w: 8, h: 0.45,
|
| 83 |
+
fontSize: 18, fontFace: BF, color: LT_BLUE, align: "center"
|
| 84 |
+
});
|
| 85 |
+
s1.addText("March 2026", {
|
| 86 |
+
x: 1, y: 4.2, w: 8, h: 0.35,
|
| 87 |
+
fontSize: 14, fontFace: BF, color: TXT_LT, align: "center"
|
| 88 |
+
});
|
| 89 |
+
|
| 90 |
+
// ════════════════════════════════════════════════════════════
|
| 91 |
+
// SLIDE 2: TASK DEFINITION
|
| 92 |
+
// ════════════════════════════════════════════════════════════
|
| 93 |
+
let s2 = pres.addSlide();
|
| 94 |
+
headerBar(s2, "Task: Single-Cell Perturbation Prediction");
|
| 95 |
+
slideNum(s2, 2);
|
| 96 |
+
|
| 97 |
+
// Left column
|
| 98 |
+
sectionLabel(s2, "Virtual Cell Vision", 0.6, 1.05);
|
| 99 |
+
s2.addText([
|
| 100 |
+
{ text: "AI model simulating cell behavior", options: { bullet: true, breakLine: true } },
|
| 101 |
+
{ text: "Predict molecular state under perturbation", options: { bullet: true, breakLine: true } },
|
| 102 |
+
{ text: "Focus: CRISPR genetic perturbation", options: { bullet: true } }
|
| 103 |
+
], {
|
| 104 |
+
x: 0.6, y: 1.5, w: 4.2, h: 1.2,
|
| 105 |
+
fontSize: 13, fontFace: BF, color: TXT, paraSpaceAfter: 4
|
| 106 |
+
});
|
| 107 |
+
|
| 108 |
+
sectionLabel(s2, "Perturbation Types", 0.6, 2.8);
|
| 109 |
+
s2.addText([
|
| 110 |
+
{ text: "Drug (small molecule compounds)", options: { bullet: true, breakLine: true } },
|
| 111 |
+
{ text: "Cytokine (immune signaling)", options: { bullet: true, breakLine: true } },
|
| 112 |
+
{ text: "Genetic (CRISPR KO / OE / KD)", options: { bullet: true, bold: true } }
|
| 113 |
+
], {
|
| 114 |
+
x: 0.6, y: 3.25, w: 4.2, h: 1.2,
|
| 115 |
+
fontSize: 13, fontFace: BF, color: TXT, paraSpaceAfter: 4
|
| 116 |
+
});
|
| 117 |
+
|
| 118 |
+
// Right column - task formulation card
|
| 119 |
+
card(s2, 5.3, 1.05, 4.2, 3.7, BLUE);
|
| 120 |
+
s2.addText("Task Formulation", {
|
| 121 |
+
x: 5.6, y: 1.15, w: 3.7, h: 0.35,
|
| 122 |
+
fontSize: 15, fontFace: HF, color: NAVY, bold: true, margin: 0
|
| 123 |
+
});
|
| 124 |
+
s2.addText([
|
| 125 |
+
{ text: "Input:", options: { bold: true, breakLine: true } },
|
| 126 |
+
{ text: " x_ctrl (control expression, G dims)", options: { breakLine: true } },
|
| 127 |
+
{ text: " p (perturbed gene ID)", options: { breakLine: true } },
|
| 128 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 129 |
+
{ text: "Output:", options: { bold: true, breakLine: true } },
|
| 130 |
+
{ text: " x_pert (perturbed expression, G dims)", options: { breakLine: true } },
|
| 131 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 132 |
+
{ text: "G = 5,000 highly variable genes", options: { italic: true, color: TXT_MID } }
|
| 133 |
+
], {
|
| 134 |
+
x: 5.6, y: 1.6, w: 3.7, h: 2.8,
|
| 135 |
+
fontSize: 12, fontFace: CF, color: TXT
|
| 136 |
+
});
|
| 137 |
+
|
| 138 |
+
// ════════════════════════════════════════════════════════════
|
| 139 |
+
// SLIDE 3: SIGNIFICANCE & DATASET
|
| 140 |
+
// ════════════════════════════════════════════════════════════
|
| 141 |
+
let s3 = pres.addSlide();
|
| 142 |
+
headerBar(s3, "Significance & Dataset");
|
| 143 |
+
slideNum(s3, 3);
|
| 144 |
+
|
| 145 |
+
// Three importance cards
|
| 146 |
+
const importCards = [
|
| 147 |
+
{ num: "$$$", title: "Drug Screening", body: "Wet-lab Perturb-seq is expensive;\nvirtual screening saves resources" },
|
| 148 |
+
{ num: "N\u00B2", title: "Combinatorial Explosion", body: "N genes \u2192 N(N-1)/2 combinations;\nimpossible to enumerate all" },
|
| 149 |
+
{ num: "DNA", title: "Disease Mechanism", body: "Predict which gene perturbation\ncauses disease phenotype" }
|
| 150 |
+
];
|
| 151 |
+
importCards.forEach((c, i) => {
|
| 152 |
+
const cx = 0.6 + i * 3.1;
|
| 153 |
+
card(s3, cx, 1.05, 2.8, 2.0, BLUE);
|
| 154 |
+
s3.addText(c.num, {
|
| 155 |
+
x: cx + 0.15, y: 1.15, w: 1.0, h: 0.45,
|
| 156 |
+
fontSize: 20, fontFace: HF, color: BLUE, bold: true, margin: 0
|
| 157 |
+
});
|
| 158 |
+
s3.addText(c.title, {
|
| 159 |
+
x: cx + 0.15, y: 1.6, w: 2.5, h: 0.3,
|
| 160 |
+
fontSize: 14, fontFace: HF, color: NAVY, bold: true, margin: 0
|
| 161 |
+
});
|
| 162 |
+
s3.addText(c.body, {
|
| 163 |
+
x: cx + 0.15, y: 1.95, w: 2.5, h: 0.9,
|
| 164 |
+
fontSize: 11, fontFace: BF, color: TXT_MID
|
| 165 |
+
});
|
| 166 |
+
});
|
| 167 |
+
|
| 168 |
+
// Dataset info
|
| 169 |
+
sectionLabel(s3, "Dataset: Norman et al.", 0.6, 3.3);
|
| 170 |
+
s3.addText([
|
| 171 |
+
{ text: "~9,000 cells \u00D7 5,000 HVG", options: { bullet: true, breakLine: true, bold: true } },
|
| 172 |
+
{ text: "105 single/double CRISPR perturbations (KO + OE)", options: { bullet: true, breakLine: true } },
|
| 173 |
+
{ text: "No cell-level pairing (destructive measurement)", options: { bullet: true, breakLine: true, bold: true, color: RED } },
|
| 174 |
+
{ text: "Metrics: DE gene overlap, direction, MSE, Pearson r", options: { bullet: true } }
|
| 175 |
+
], {
|
| 176 |
+
x: 0.6, y: 3.75, w: 8.8, h: 1.3,
|
| 177 |
+
fontSize: 13, fontFace: BF, color: TXT, paraSpaceAfter: 4
|
| 178 |
+
});
|
| 179 |
+
|
| 180 |
+
// ════════════════════════════════════════════════════════════
|
| 181 |
+
// SLIDE 4: EXISTING METHODS
|
| 182 |
+
// ════════════════════════════════════════════════════════════
|
| 183 |
+
let s4 = pres.addSlide();
|
| 184 |
+
headerBar(s4, "Existing Methods & Limitations");
|
| 185 |
+
slideNum(s4, 4);
|
| 186 |
+
|
| 187 |
+
const hdrOpts = (txt) => ({ text: txt, options: { bold: true, color: WHITE, fill: { color: NAVY }, fontSize: 12, fontFace: BF, align: "center" } });
|
| 188 |
+
const cellOpts = (txt, opts) => ({ text: txt, options: { fontSize: 11, fontFace: BF, ...opts } });
|
| 189 |
+
const altBg = { fill: { color: "F8F9FA" } };
|
| 190 |
+
|
| 191 |
+
s4.addTable([
|
| 192 |
+
[ hdrOpts("Method"), hdrOpts("Type"), hdrOpts("Key Limitation") ],
|
| 193 |
+
[ cellOpts("Additive Shift", altBg), cellOpts("Baseline", { ...altBg, align: "center" }), cellOpts("Ignores cell heterogeneity; constant shift assumption", altBg) ],
|
| 194 |
+
[ cellOpts("scGPT"), cellOpts("Pretrained LM", { align: "center" }), cellOpts("Autoregressive completion; not designed for perturbation") ],
|
| 195 |
+
[ cellOpts("Geneformer", altBg), cellOpts("Pretrained LM", { ...altBg, align: "center" }), cellOpts("Heuristic in-silico perturbation; loses expression info", altBg) ],
|
| 196 |
+
[ cellOpts("CPA"), cellOpts("Specialized", { align: "center" }), cellOpts("Linear additivity assumption in latent space") ],
|
| 197 |
+
[ cellOpts("GEARS", altBg), cellOpts("Specialized", { ...altBg, align: "center" }), cellOpts("Static GO graph prior; deterministic prediction only", altBg) ],
|
| 198 |
+
[ cellOpts("scDFM", { bold: true }), cellOpts("Flow Matching", { align: "center" }), cellOpts("No GRN modeling; limited model capacity (d=128)") ]
|
| 199 |
+
], {
|
| 200 |
+
x: 0.6, y: 1.1, w: 8.8,
|
| 201 |
+
colW: [1.8, 1.5, 5.5],
|
| 202 |
+
border: { pt: 0.5, color: "DDE1E6" },
|
| 203 |
+
rowH: [0.45, 0.42, 0.42, 0.42, 0.42, 0.42, 0.42]
|
| 204 |
+
});
|
| 205 |
+
|
| 206 |
+
s4.addText("scDFM (ICLR 2026) is closest to our work \u2014 we build upon its flow matching framework.", {
|
| 207 |
+
x: 0.6, y: 4.5, w: 8.8, h: 0.3,
|
| 208 |
+
fontSize: 11, fontFace: BF, color: TXT_MID, italic: true
|
| 209 |
+
});
|
| 210 |
+
|
| 211 |
+
// ════════════════════════════════════════════════════════════
|
| 212 |
+
// SLIDE 5: THE MISSING PIECE
|
| 213 |
+
// ════════════════════════════════════════════════════════════
|
| 214 |
+
let s5 = pres.addSlide();
|
| 215 |
+
headerBar(s5, "The Common Blind Spot");
|
| 216 |
+
slideNum(s5, 5);
|
| 217 |
+
|
| 218 |
+
sectionLabel(s5, "All existing methods share the same gap:");
|
| 219 |
+
|
| 220 |
+
// Existing approach block
|
| 221 |
+
card(s5, 0.6, 1.8, 8.8, 1.3, RED);
|
| 222 |
+
s5.addText("Existing Approach", {
|
| 223 |
+
x: 0.85, y: 1.9, w: 3.0, h: 0.3,
|
| 224 |
+
fontSize: 14, fontFace: HF, color: RED, bold: true, margin: 0
|
| 225 |
+
});
|
| 226 |
+
s5.addText("Perturbation \u2192 [ Black-Box Model ] \u2192 Expression Change", {
|
| 227 |
+
x: 0.85, y: 2.3, w: 8.2, h: 0.4,
|
| 228 |
+
fontSize: 16, fontFace: CF, color: TXT, margin: 0
|
| 229 |
+
});
|
| 230 |
+
s5.addText("No explicit modeling of gene regulatory network changes", {
|
| 231 |
+
x: 0.85, y: 2.7, w: 8.0, h: 0.3,
|
| 232 |
+
fontSize: 12, fontFace: BF, color: TXT_MID, italic: true, margin: 0
|
| 233 |
+
});
|
| 234 |
+
|
| 235 |
+
// Our approach block
|
| 236 |
+
card(s5, 0.6, 3.5, 8.8, 1.3, GREEN);
|
| 237 |
+
s5.addText("Our Approach", {
|
| 238 |
+
x: 0.85, y: 3.6, w: 3.0, h: 0.3,
|
| 239 |
+
fontSize: 14, fontFace: HF, color: GREEN, bold: true, margin: 0
|
| 240 |
+
});
|
| 241 |
+
s5.addText("Perturbation \u2192 GRN Rewiring \u2192 Expression Change", {
|
| 242 |
+
x: 0.85, y: 4.0, w: 8.2, h: 0.4,
|
| 243 |
+
fontSize: 16, fontFace: CF, color: TXT, margin: 0
|
| 244 |
+
});
|
| 245 |
+
s5.addText("Explicitly model how perturbation alters the gene regulatory network", {
|
| 246 |
+
x: 0.85, y: 4.4, w: 8.0, h: 0.3,
|
| 247 |
+
fontSize: 12, fontFace: BF, color: TXT_MID, italic: true, margin: 0
|
| 248 |
+
});
|
| 249 |
+
|
| 250 |
+
// ════════════════════════════════════════════════════════════
|
| 251 |
+
// SLIDE 6: THREE KEY MOTIVATIONS
|
| 252 |
+
// ════════════════════════════════════════════════════════════
|
| 253 |
+
let s6 = pres.addSlide();
|
| 254 |
+
headerBar(s6, "Three Key Motivations");
|
| 255 |
+
slideNum(s6, 6);
|
| 256 |
+
|
| 257 |
+
const motivations = [
|
| 258 |
+
{
|
| 259 |
+
num: "1", title: "Flow Matching for Unpaired Data",
|
| 260 |
+
bullets: [
|
| 261 |
+
"Learns probability transport: p(ctrl) \u2192 p(pert)",
|
| 262 |
+
"No cell-level pairing required",
|
| 263 |
+
"Generative output with uncertainty estimation"
|
| 264 |
+
]
|
| 265 |
+
},
|
| 266 |
+
{
|
| 267 |
+
num: "2", title: "GRN Cascade Drives Expression Change",
|
| 268 |
+
bullets: [
|
| 269 |
+
"KO gene A \u2192 direct targets B,C,D change",
|
| 270 |
+
"Cascade propagates through regulatory network",
|
| 271 |
+
"Understanding GRN change = better prediction"
|
| 272 |
+
]
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
num: "3", title: "scGPT Attention \u2248 Data-Driven GRN",
|
| 276 |
+
bullets: [
|
| 277 |
+
"Pretrained attention encodes gene-gene regulation",
|
| 278 |
+
"Context-dependent: varies with cell state",
|
| 279 |
+
"\u0394_attn captures GRN change from perturbation"
|
| 280 |
+
]
|
| 281 |
+
}
|
| 282 |
+
];
|
| 283 |
+
|
| 284 |
+
motivations.forEach((m, i) => {
|
| 285 |
+
const cy = 1.05 + i * 1.4;
|
| 286 |
+
card(s6, 0.6, cy, 8.8, 1.2, BLUE);
|
| 287 |
+
s6.addShape(pres.shapes.OVAL, {
|
| 288 |
+
x: 0.85, y: cy + 0.15, w: 0.5, h: 0.5,
|
| 289 |
+
fill: { color: NAVY }
|
| 290 |
+
});
|
| 291 |
+
s6.addText(m.num, {
|
| 292 |
+
x: 0.85, y: cy + 0.15, w: 0.5, h: 0.5,
|
| 293 |
+
fontSize: 18, fontFace: HF, color: WHITE, bold: true,
|
| 294 |
+
align: "center", valign: "middle", margin: 0
|
| 295 |
+
});
|
| 296 |
+
s6.addText(m.title, {
|
| 297 |
+
x: 1.55, y: cy + 0.1, w: 7.5, h: 0.35,
|
| 298 |
+
fontSize: 15, fontFace: HF, color: NAVY, bold: true, margin: 0
|
| 299 |
+
});
|
| 300 |
+
s6.addText(m.bullets.map((b, bi) => ({
|
| 301 |
+
text: b,
|
| 302 |
+
options: { bullet: true, breakLine: bi < m.bullets.length - 1 }
|
| 303 |
+
})), {
|
| 304 |
+
x: 1.55, y: cy + 0.5, w: 7.5, h: 0.65,
|
| 305 |
+
fontSize: 12, fontFace: BF, color: TXT_MID, paraSpaceAfter: 2
|
| 306 |
+
});
|
| 307 |
+
});
|
| 308 |
+
|
| 309 |
+
// ════════════════════════════════════════════════════════════
|
| 310 |
+
// SLIDE 7: METHOD OVERVIEW
|
| 311 |
+
// ════════════════════════════════════════════════════════════
|
| 312 |
+
let s7 = pres.addSlide();
|
| 313 |
+
headerBar(s7, "Method: Cascaded Flow Matching");
|
| 314 |
+
slideNum(s7, 7);
|
| 315 |
+
|
| 316 |
+
sectionLabel(s7, "Two-Stage Generation: \"Think First, Then Predict\"");
|
| 317 |
+
|
| 318 |
+
// Stage 1 box
|
| 319 |
+
card(s7, 0.6, 1.7, 4.1, 2.8, ORANGE);
|
| 320 |
+
s7.addText("Stage 1: GRN Latent Flow", {
|
| 321 |
+
x: 0.85, y: 1.8, w: 3.6, h: 0.35,
|
| 322 |
+
fontSize: 15, fontFace: HF, color: ORANGE, bold: true, margin: 0
|
| 323 |
+
});
|
| 324 |
+
s7.addText([
|
| 325 |
+
{ text: "noise \u2192 GRN change features", options: { breakLine: true, bold: true } },
|
| 326 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 327 |
+
{ text: "Understand how perturbation", options: { breakLine: true } },
|
| 328 |
+
{ text: "rewires the regulatory network", options: { breakLine: true } },
|
| 329 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 330 |
+
{ text: "Conditioned on:", options: { bold: true, breakLine: true } },
|
| 331 |
+
{ text: " \u2022 control expression", options: { breakLine: true } },
|
| 332 |
+
{ text: " \u2022 perturbed gene ID", options: {} }
|
| 333 |
+
], {
|
| 334 |
+
x: 0.85, y: 2.25, w: 3.6, h: 2.0,
|
| 335 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 336 |
+
});
|
| 337 |
+
|
| 338 |
+
// Arrow
|
| 339 |
+
s7.addText("\u2192", {
|
| 340 |
+
x: 4.7, y: 2.7, w: 0.6, h: 0.5,
|
| 341 |
+
fontSize: 30, fontFace: BF, color: NAVY, align: "center", valign: "middle", bold: true
|
| 342 |
+
});
|
| 343 |
+
|
| 344 |
+
// Stage 2 box
|
| 345 |
+
card(s7, 5.3, 1.7, 4.1, 2.8, GREEN);
|
| 346 |
+
s7.addText("Stage 2: Expression Flow", {
|
| 347 |
+
x: 5.55, y: 1.8, w: 3.6, h: 0.35,
|
| 348 |
+
fontSize: 15, fontFace: HF, color: GREEN, bold: true, margin: 0
|
| 349 |
+
});
|
| 350 |
+
s7.addText([
|
| 351 |
+
{ text: "noise \u2192 gene expression", options: { breakLine: true, bold: true } },
|
| 352 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 353 |
+
{ text: "Predict expression changes", options: { breakLine: true } },
|
| 354 |
+
{ text: "based on GRN understanding", options: { breakLine: true } },
|
| 355 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 356 |
+
{ text: "Conditioned on:", options: { bold: true, breakLine: true } },
|
| 357 |
+
{ text: " \u2022 Stage 1 GRN features", options: { breakLine: true } },
|
| 358 |
+
{ text: " \u2022 control expression + pert ID", options: {} }
|
| 359 |
+
], {
|
| 360 |
+
x: 5.55, y: 2.25, w: 3.6, h: 2.0,
|
| 361 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 362 |
+
});
|
| 363 |
+
|
| 364 |
+
// Bottom insight bar
|
| 365 |
+
s7.addShape(pres.shapes.RECTANGLE, {
|
| 366 |
+
x: 0.6, y: 4.7, w: 8.8, h: 0.5,
|
| 367 |
+
fill: { color: LT_BLUE }
|
| 368 |
+
});
|
| 369 |
+
s7.addText("Biological intuition: first understand GRN rewiring, then predict expression change", {
|
| 370 |
+
x: 0.8, y: 4.7, w: 8.4, h: 0.5,
|
| 371 |
+
fontSize: 13, fontFace: BF, color: NAVY, italic: true, valign: "middle"
|
| 372 |
+
});
|
| 373 |
+
|
| 374 |
+
// ════════════════════════════════════════════════════════════
|
| 375 |
+
// SLIDE 8: GRN FEATURE EXTRACTION
|
| 376 |
+
// ════════════════════════════════════════════════════════════
|
| 377 |
+
let s8 = pres.addSlide();
|
| 378 |
+
headerBar(s8, "GRN Feature: Attention-Delta Extraction");
|
| 379 |
+
slideNum(s8, 8);
|
| 380 |
+
|
| 381 |
+
sectionLabel(s8, "Using Frozen scGPT to Extract GRN Change Signal");
|
| 382 |
+
|
| 383 |
+
// Steps
|
| 384 |
+
const steps = [
|
| 385 |
+
"Feed control & perturbed expression\ninto frozen scGPT separately",
|
| 386 |
+
"Extract attention matrices:\nAttn(ctrl) and Attn(pert)",
|
| 387 |
+
"Compute delta:\n\u0394_attn = Attn(pert) \u2212 Attn(ctrl)",
|
| 388 |
+
"Project to features:\nz = \u0394_attn \u00D7 gene_embeddings"
|
| 389 |
+
];
|
| 390 |
+
steps.forEach((desc, i) => {
|
| 391 |
+
const sy = 1.65 + i * 0.9;
|
| 392 |
+
s8.addShape(pres.shapes.OVAL, {
|
| 393 |
+
x: 0.7, y: sy + 0.05, w: 0.45, h: 0.45,
|
| 394 |
+
fill: { color: BLUE }
|
| 395 |
+
});
|
| 396 |
+
s8.addText(String(i + 1), {
|
| 397 |
+
x: 0.7, y: sy + 0.05, w: 0.45, h: 0.45,
|
| 398 |
+
fontSize: 16, fontFace: HF, color: WHITE, bold: true,
|
| 399 |
+
align: "center", valign: "middle", margin: 0
|
| 400 |
+
});
|
| 401 |
+
s8.addText(desc, {
|
| 402 |
+
x: 1.4, y: sy, w: 3.8, h: 0.6,
|
| 403 |
+
fontSize: 12, fontFace: BF, color: TXT, valign: "middle", margin: 0
|
| 404 |
+
});
|
| 405 |
+
if (i < steps.length - 1) {
|
| 406 |
+
s8.addShape(pres.shapes.LINE, {
|
| 407 |
+
x: 0.92, y: sy + 0.52, w: 0, h: 0.35,
|
| 408 |
+
line: { color: BLUE, width: 1.5, dashType: "dash" }
|
| 409 |
+
});
|
| 410 |
+
}
|
| 411 |
+
});
|
| 412 |
+
|
| 413 |
+
// Output card on right
|
| 414 |
+
card(s8, 5.6, 1.65, 3.8, 3.2, NAVY);
|
| 415 |
+
s8.addText("Output", {
|
| 416 |
+
x: 5.85, y: 1.75, w: 3.3, h: 0.3,
|
| 417 |
+
fontSize: 15, fontFace: HF, color: NAVY, bold: true, margin: 0
|
| 418 |
+
});
|
| 419 |
+
s8.addText([
|
| 420 |
+
{ text: "Per-gene GRN change vector", options: { breakLine: true, bold: true } },
|
| 421 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 422 |
+
{ text: "Shape: (B, G, 512)", options: { breakLine: true, fontFace: CF } },
|
| 423 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 424 |
+
{ text: "Each gene gets a 512-d vector\nencoding: \"how did upstream\nregulatory relationships change\nfor this gene?\"", options: { color: TXT_MID } }
|
| 425 |
+
], {
|
| 426 |
+
x: 5.85, y: 2.15, w: 3.3, h: 2.2,
|
| 427 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 428 |
+
});
|
| 429 |
+
|
| 430 |
+
// ════════════════════════════════════════════════════════════
|
| 431 |
+
// SLIDE 9: MODEL ARCHITECTURE
|
| 432 |
+
// ════════════════════════════════════════════════════════════
|
| 433 |
+
let s9 = pres.addSlide();
|
| 434 |
+
headerBar(s9, "Model Architecture");
|
| 435 |
+
slideNum(s9, 9);
|
| 436 |
+
|
| 437 |
+
// Expression Stream
|
| 438 |
+
card(s9, 0.6, 1.1, 4.1, 1.1, BLUE);
|
| 439 |
+
s9.addText("Expression Stream", {
|
| 440 |
+
x: 0.85, y: 1.15, w: 3.6, h: 0.3,
|
| 441 |
+
fontSize: 14, fontFace: HF, color: BLUE, bold: true, margin: 0
|
| 442 |
+
});
|
| 443 |
+
s9.addText("GeneEncoder + ValueEnc \u2192 expr_tokens (B,G,d)", {
|
| 444 |
+
x: 0.85, y: 1.5, w: 3.6, h: 0.4,
|
| 445 |
+
fontSize: 11, fontFace: CF, color: TXT, margin: 0
|
| 446 |
+
});
|
| 447 |
+
|
| 448 |
+
// Latent Stream
|
| 449 |
+
card(s9, 5.3, 1.1, 4.1, 1.1, ORANGE);
|
| 450 |
+
s9.addText("Latent Stream", {
|
| 451 |
+
x: 5.55, y: 1.15, w: 3.6, h: 0.3,
|
| 452 |
+
fontSize: 14, fontFace: HF, color: ORANGE, bold: true, margin: 0
|
| 453 |
+
});
|
| 454 |
+
s9.addText("LatentEmbedder \u2192 lat_tokens (B,G,d)", {
|
| 455 |
+
x: 5.55, y: 1.5, w: 3.6, h: 0.4,
|
| 456 |
+
fontSize: 11, fontFace: CF, color: TXT, margin: 0
|
| 457 |
+
});
|
| 458 |
+
|
| 459 |
+
// Merge
|
| 460 |
+
s9.addText("\u2295 Additive Fusion", {
|
| 461 |
+
x: 3.0, y: 2.35, w: 4.0, h: 0.35,
|
| 462 |
+
fontSize: 13, fontFace: BF, color: NAVY, bold: true, align: "center", margin: 0
|
| 463 |
+
});
|
| 464 |
+
|
| 465 |
+
// Down arrows
|
| 466 |
+
s9.addShape(pres.shapes.LINE, {
|
| 467 |
+
x: 2.5, y: 2.2, w: 0, h: 0.15,
|
| 468 |
+
line: { color: NAVY, width: 1.5 }
|
| 469 |
+
});
|
| 470 |
+
s9.addShape(pres.shapes.LINE, {
|
| 471 |
+
x: 7.5, y: 2.2, w: 0, h: 0.15,
|
| 472 |
+
line: { color: NAVY, width: 1.5 }
|
| 473 |
+
});
|
| 474 |
+
|
| 475 |
+
// Conditioning
|
| 476 |
+
card(s9, 2.0, 2.85, 6.0, 0.5, NAVY);
|
| 477 |
+
s9.addText("Conditioning: c = t_expr + t_latent + pert_embedding", {
|
| 478 |
+
x: 2.25, y: 2.9, w: 5.5, h: 0.4,
|
| 479 |
+
fontSize: 11, fontFace: CF, color: TXT, valign: "middle", margin: 0
|
| 480 |
+
});
|
| 481 |
+
|
| 482 |
+
// Down arrow to backbone
|
| 483 |
+
s9.addShape(pres.shapes.LINE, {
|
| 484 |
+
x: 5.0, y: 3.35, w: 0, h: 0.2,
|
| 485 |
+
line: { color: NAVY, width: 1.5 }
|
| 486 |
+
});
|
| 487 |
+
|
| 488 |
+
// Shared Backbone
|
| 489 |
+
s9.addShape(pres.shapes.RECTANGLE, {
|
| 490 |
+
x: 2.0, y: 3.6, w: 6.0, h: 0.65,
|
| 491 |
+
fill: { color: NAVY }
|
| 492 |
+
});
|
| 493 |
+
s9.addText("Shared Backbone: DiffPerceiverBlock \u00D7 4 (with Gene-AdaLN)", {
|
| 494 |
+
x: 2.0, y: 3.6, w: 6.0, h: 0.65,
|
| 495 |
+
fontSize: 13, fontFace: BF, color: WHITE, bold: true,
|
| 496 |
+
align: "center", valign: "middle"
|
| 497 |
+
});
|
| 498 |
+
|
| 499 |
+
// Down arrows to heads
|
| 500 |
+
s9.addShape(pres.shapes.LINE, {
|
| 501 |
+
x: 3.4, y: 4.25, w: 0, h: 0.2,
|
| 502 |
+
line: { color: NAVY, width: 1.5 }
|
| 503 |
+
});
|
| 504 |
+
s9.addShape(pres.shapes.LINE, {
|
| 505 |
+
x: 6.6, y: 4.25, w: 0, h: 0.2,
|
| 506 |
+
line: { color: NAVY, width: 1.5 }
|
| 507 |
+
});
|
| 508 |
+
|
| 509 |
+
// Dual heads
|
| 510 |
+
card(s9, 2.0, 4.5, 2.8, 0.65, BLUE);
|
| 511 |
+
s9.addText("Expression Head \u2192 v_expr (B,G)", {
|
| 512 |
+
x: 2.2, y: 4.55, w: 2.4, h: 0.45,
|
| 513 |
+
fontSize: 11, fontFace: CF, color: TXT, valign: "middle", margin: 0
|
| 514 |
+
});
|
| 515 |
+
card(s9, 5.2, 4.5, 2.8, 0.65, ORANGE);
|
| 516 |
+
s9.addText("Latent Head \u2192 v_latent (B,G,512)", {
|
| 517 |
+
x: 5.4, y: 4.55, w: 2.4, h: 0.45,
|
| 518 |
+
fontSize: 11, fontFace: CF, color: TXT, valign: "middle", margin: 0
|
| 519 |
+
});
|
| 520 |
+
|
| 521 |
+
// ════════════════════════════════════════════════════════════
|
| 522 |
+
// SLIDE 10: TRAINING & INFERENCE
|
| 523 |
+
// ════════════════════════════════════════════════════════════
|
| 524 |
+
let s10 = pres.addSlide();
|
| 525 |
+
headerBar(s10, "Cascaded Training & Inference");
|
| 526 |
+
slideNum(s10, 10);
|
| 527 |
+
|
| 528 |
+
// Left: Training
|
| 529 |
+
sectionLabel(s10, "Training: Probabilistic Switching", 0.6, 1.1, 4.2);
|
| 530 |
+
card(s10, 0.6, 1.55, 4.2, 1.4, BLUE);
|
| 531 |
+
s10.addText([
|
| 532 |
+
{ text: "40%", options: { bold: true, fontSize: 20, color: ORANGE } },
|
| 533 |
+
{ text: " Train Latent Flow only", options: { fontSize: 13 } }
|
| 534 |
+
], {
|
| 535 |
+
x: 0.85, y: 1.65, w: 3.7, h: 0.45, fontFace: BF, color: TXT, valign: "middle", margin: 0
|
| 536 |
+
});
|
| 537 |
+
s10.addText("t\u2082 random, t\u2081 = 0, only loss_latent", {
|
| 538 |
+
x: 0.85, y: 2.05, w: 3.7, h: 0.25,
|
| 539 |
+
fontSize: 10, fontFace: CF, color: TXT_MID, margin: 0
|
| 540 |
+
});
|
| 541 |
+
s10.addText([
|
| 542 |
+
{ text: "60%", options: { bold: true, fontSize: 20, color: GREEN } },
|
| 543 |
+
{ text: " Train Expression Flow only", options: { fontSize: 13 } }
|
| 544 |
+
], {
|
| 545 |
+
x: 0.85, y: 2.4, w: 3.7, h: 0.45, fontFace: BF, color: TXT, valign: "middle", margin: 0
|
| 546 |
+
});
|
| 547 |
+
s10.addText("t\u2081 random, t\u2082 \u2248 1, only loss_expr", {
|
| 548 |
+
x: 0.85, y: 2.7, w: 3.7, h: 0.25,
|
| 549 |
+
fontSize: 10, fontFace: CF, color: TXT_MID, margin: 0
|
| 550 |
+
});
|
| 551 |
+
|
| 552 |
+
// Right: Inference
|
| 553 |
+
sectionLabel(s10, "Inference: Sequential Two-Stage", 5.3, 1.1, 4.2);
|
| 554 |
+
card(s10, 5.3, 1.55, 4.2, 1.4, NAVY);
|
| 555 |
+
s10.addText([
|
| 556 |
+
{ text: "Stage 1:", options: { bold: true, color: ORANGE, breakLine: true } },
|
| 557 |
+
{ text: "z_noise \u2550\u2550(ODE)\u2550\u2550> z_clean (t\u2082: 0\u21921)", options: { fontFace: CF, fontSize: 11, breakLine: true } },
|
| 558 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 559 |
+
{ text: "Stage 2:", options: { bold: true, color: GREEN, breakLine: true } },
|
| 560 |
+
{ text: "x_noise \u2550\u2550(ODE)\u2550\u2550> x_pred (t\u2081: 0\u21921)", options: { fontFace: CF, fontSize: 11 } }
|
| 561 |
+
], {
|
| 562 |
+
x: 5.55, y: 1.65, w: 3.7, h: 1.2,
|
| 563 |
+
fontSize: 12, fontFace: BF, color: TXT, margin: 0
|
| 564 |
+
});
|
| 565 |
+
|
| 566 |
+
// Biological analogy
|
| 567 |
+
sectionLabel(s10, "Biological Cascade Analogy", 0.6, 3.2, 8.8);
|
| 568 |
+
s10.addShape(pres.shapes.RECTANGLE, {
|
| 569 |
+
x: 0.6, y: 3.6, w: 8.8, h: 1.6,
|
| 570 |
+
fill: { color: LT_BLUE }
|
| 571 |
+
});
|
| 572 |
+
s10.addText([
|
| 573 |
+
{ text: "CRISPR knock-out gene A", options: { bold: true, breakLine: true } },
|
| 574 |
+
{ text: " \u2193 Gene A expression \u2192 0", options: { breakLine: true } },
|
| 575 |
+
{ text: " \u2193 Direct targets B, C, D change (1st-order)", options: { breakLine: true } },
|
| 576 |
+
{ text: " \u2193 B\u2019s targets E, F and C\u2019s targets G, H change (cascade)", options: { breakLine: true } },
|
| 577 |
+
{ text: " \u2193 Thousands of genes altered across the transcriptome", options: {} }
|
| 578 |
+
], {
|
| 579 |
+
x: 0.8, y: 3.65, w: 8.4, h: 1.5,
|
| 580 |
+
fontSize: 12, fontFace: CF, color: TXT
|
| 581 |
+
});
|
| 582 |
+
|
| 583 |
+
// ════════════════════════════════════════════════════════════
|
| 584 |
+
// SLIDE 11: CHALLENGE 1 - NOISY GRN SIGNAL
|
| 585 |
+
// ════════════════════════════════════════════════════════════
|
| 586 |
+
let s11 = pres.addSlide();
|
| 587 |
+
headerBar(s11, "Challenge 1: Noisy GRN Signal");
|
| 588 |
+
slideNum(s11, 11);
|
| 589 |
+
|
| 590 |
+
// Problem
|
| 591 |
+
sectionLabel(s11, "Problem: Noise Drowns True Signal", 0.6, 1.1, 4.2);
|
| 592 |
+
card(s11, 0.6, 1.55, 4.2, 1.8, RED);
|
| 593 |
+
s11.addText([
|
| 594 |
+
{ text: "Attention matrix: 5000\u00D75000", options: { bold: true, breakLine: true } },
|
| 595 |
+
{ text: "= 25,000,000 non-zero values", options: { breakLine: true } },
|
| 596 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 597 |
+
{ text: "Real GRN: ~20\u201350 targets per gene", options: { breakLine: true } },
|
| 598 |
+
{ text: "\u2192 99%+ attention values are noise", options: { bold: true, color: RED } }
|
| 599 |
+
], {
|
| 600 |
+
x: 0.85, y: 1.65, w: 3.7, h: 1.4,
|
| 601 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 602 |
+
});
|
| 603 |
+
s11.addText("Evidence: latent loss \u2248 1.12 >> expr loss \u2248 0.019", {
|
| 604 |
+
x: 0.6, y: 3.5, w: 4.2, h: 0.25,
|
| 605 |
+
fontSize: 11, fontFace: BF, color: TXT_MID, italic: true
|
| 606 |
+
});
|
| 607 |
+
|
| 608 |
+
// Solution
|
| 609 |
+
sectionLabel(s11, "Solution: Sparse Top-K Filtering", 5.3, 1.1, 4.2);
|
| 610 |
+
card(s11, 5.3, 1.55, 4.2, 1.8, GREEN);
|
| 611 |
+
s11.addText([
|
| 612 |
+
{ text: "Keep only top K=30 per gene", options: { bold: true, breakLine: true } },
|
| 613 |
+
{ text: "(ranked by |\u0394_attn| magnitude)", options: { breakLine: true, color: TXT_MID } },
|
| 614 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 615 |
+
{ text: "\u2192 Filters 99.4% noise", options: { bold: true, color: GREEN, breakLine: true } },
|
| 616 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 617 |
+
{ text: "features = sparse_\u0394_topk \u00D7 gene_emb", options: { fontFace: CF, fontSize: 11 } }
|
| 618 |
+
], {
|
| 619 |
+
x: 5.55, y: 1.65, w: 3.7, h: 1.4,
|
| 620 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 621 |
+
});
|
| 622 |
+
s11.addText("Status: implemented (sparse_topk_emb mode)", {
|
| 623 |
+
x: 5.3, y: 3.5, w: 4.2, h: 0.25,
|
| 624 |
+
fontSize: 11, fontFace: BF, color: GREEN, italic: true
|
| 625 |
+
});
|
| 626 |
+
|
| 627 |
+
// Before/after comparison bar
|
| 628 |
+
s11.addShape(pres.shapes.RECTANGLE, {
|
| 629 |
+
x: 0.6, y: 4.0, w: 8.8, h: 1.2,
|
| 630 |
+
fill: { color: BG_GRAY }
|
| 631 |
+
});
|
| 632 |
+
s11.addText([
|
| 633 |
+
{ text: "Before: ", options: { bold: true } },
|
| 634 |
+
{ text: "\u0394_attn (G\u00D7G) \u2192 25M values \u2192 noise dominates \u2192 loss ~1.12", options: { color: RED } }
|
| 635 |
+
], {
|
| 636 |
+
x: 0.8, y: 4.1, w: 8.4, h: 0.35,
|
| 637 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 638 |
+
});
|
| 639 |
+
s11.addText([
|
| 640 |
+
{ text: "After: ", options: { bold: true } },
|
| 641 |
+
{ text: "sparse_\u0394_topk (G\u00D7K) \u2192 150K values \u2192 signal preserved \u2192 loss expected \u2193", options: { color: GREEN } }
|
| 642 |
+
], {
|
| 643 |
+
x: 0.8, y: 4.55, w: 8.4, h: 0.35,
|
| 644 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 645 |
+
});
|
| 646 |
+
|
| 647 |
+
// ════════════════════════════════════════════════════════════
|
| 648 |
+
// SLIDE 12: CHALLENGE 2 - HIGH-DIM LATENT
|
| 649 |
+
// ════════════════════════════════════════════════════════════
|
| 650 |
+
let s12 = pres.addSlide();
|
| 651 |
+
headerBar(s12, "Challenge 2: High-Dimensional Latent");
|
| 652 |
+
slideNum(s12, 12);
|
| 653 |
+
|
| 654 |
+
// Problem
|
| 655 |
+
sectionLabel(s12, "Problem: High-Dim Latent Prediction", 0.6, 1.1, 4.2);
|
| 656 |
+
card(s12, 0.6, 1.55, 4.2, 1.8, RED);
|
| 657 |
+
s12.addText([
|
| 658 |
+
{ text: "Each gene: 512-d GRN feature vector", options: { breakLine: true, bold: true } },
|
| 659 |
+
{ text: "Total: G\u00D7512 = 2.5M-dim velocity field", options: { breakLine: true } },
|
| 660 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 661 |
+
{ text: "Ablation experiment:", options: { bold: true, breakLine: true } },
|
| 662 |
+
{ text: "512-d \u2192 1-d: loss drops 1.1 \u2192 0.5", options: { bold: true, color: RED } }
|
| 663 |
+
], {
|
| 664 |
+
x: 0.85, y: 1.65, w: 3.7, h: 1.5,
|
| 665 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 666 |
+
});
|
| 667 |
+
s12.addText("Dimensionality is a major difficulty source", {
|
| 668 |
+
x: 0.6, y: 3.5, w: 4.2, h: 0.25,
|
| 669 |
+
fontSize: 11, fontFace: BF, color: TXT_MID, italic: true
|
| 670 |
+
});
|
| 671 |
+
|
| 672 |
+
// Solution
|
| 673 |
+
sectionLabel(s12, "Solution: PCA Compression", 5.3, 1.1, 4.2);
|
| 674 |
+
card(s12, 5.3, 1.55, 4.2, 1.8, GREEN);
|
| 675 |
+
s12.addText([
|
| 676 |
+
{ text: "PCA on gene embeddings:", options: { bold: true, breakLine: true } },
|
| 677 |
+
{ text: "512-d \u2192 64-d principal components", options: { breakLine: true } },
|
| 678 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 679 |
+
{ text: "features = sparse_\u0394 \u00D7 pca_basis", options: { fontFace: CF, fontSize: 11, breakLine: true } },
|
| 680 |
+
{ text: "Output: (B, G, 64)", options: { fontFace: CF, fontSize: 11 } }
|
| 681 |
+
], {
|
| 682 |
+
x: 5.55, y: 1.65, w: 3.7, h: 1.5,
|
| 683 |
+
fontSize: 12, fontFace: BF, color: TXT
|
| 684 |
+
});
|
| 685 |
+
s12.addText("Status: implemented (sparse_pca mode)", {
|
| 686 |
+
x: 5.3, y: 3.5, w: 4.2, h: 0.25,
|
| 687 |
+
fontSize: 11, fontFace: BF, color: GREEN, italic: true
|
| 688 |
+
});
|
| 689 |
+
|
| 690 |
+
// Combined pipeline
|
| 691 |
+
s12.addShape(pres.shapes.RECTANGLE, {
|
| 692 |
+
x: 0.6, y: 4.0, w: 8.8, h: 1.2,
|
| 693 |
+
fill: { color: LT_BLUE }
|
| 694 |
+
});
|
| 695 |
+
s12.addText("Combined Pipeline", {
|
| 696 |
+
x: 0.8, y: 4.05, w: 8.4, h: 0.3,
|
| 697 |
+
fontSize: 14, fontFace: HF, color: NAVY, bold: true, margin: 0
|
| 698 |
+
});
|
| 699 |
+
s12.addText("\u0394_attn \u2192 Sparse Top-K (noise filter) \u2192 PCA 512\u219264 (dim reduction) \u2192 GRN features (B, G, 64)", {
|
| 700 |
+
x: 0.8, y: 4.45, w: 8.4, h: 0.5,
|
| 701 |
+
fontSize: 13, fontFace: CF, color: TXT, valign: "middle"
|
| 702 |
+
});
|
| 703 |
+
|
| 704 |
+
// ════════════════════════════════════════════════════════════
|
| 705 |
+
// SLIDE 13: SUMMARY & FUTURE WORK
|
| 706 |
+
// ════════════════════════════════════════════════════════════
|
| 707 |
+
let s13 = pres.addSlide();
|
| 708 |
+
headerBar(s13, "Summary & Future Work");
|
| 709 |
+
slideNum(s13, 13);
|
| 710 |
+
|
| 711 |
+
// Core Contribution
|
| 712 |
+
sectionLabel(s13, "Core Contribution", 0.6, 1.1);
|
| 713 |
+
s13.addText([
|
| 714 |
+
{ text: "First explicit GRN modeling in perturbation prediction", options: { bullet: true, breakLine: true, bold: true } },
|
| 715 |
+
{ text: "Cascaded flow matching: GRN first, expression second", options: { bullet: true, breakLine: true } },
|
| 716 |
+
{ text: "Biologically grounded: perturbation cascades through GRN", options: { bullet: true } }
|
| 717 |
+
], {
|
| 718 |
+
x: 0.6, y: 1.5, w: 8.8, h: 1.0,
|
| 719 |
+
fontSize: 13, fontFace: BF, color: TXT, paraSpaceAfter: 4
|
| 720 |
+
});
|
| 721 |
+
|
| 722 |
+
// Future experiment
|
| 723 |
+
sectionLabel(s13, "Key Future Experiment: Validate Causal Hypothesis", 0.6, 2.7);
|
| 724 |
+
|
| 725 |
+
const fHdr = (t) => ({ text: t, options: { bold: true, color: WHITE, fill: { color: NAVY }, fontSize: 11, fontFace: BF, align: "center" } });
|
| 726 |
+
const fCell = (t, opts) => ({ text: t, options: { fontSize: 11, fontFace: BF, ...opts } });
|
| 727 |
+
const fAlt = { fill: { color: "F8F9FA" } };
|
| 728 |
+
|
| 729 |
+
s13.addTable([
|
| 730 |
+
[ fHdr("Inference Order"), fHdr("Meaning"), fHdr("Expected") ],
|
| 731 |
+
[ fCell("GRN \u2192 Expression", { ...fAlt, bold: true }), fCell("Understand first, then predict", fAlt), fCell("Best", { ...fAlt, bold: true, color: GREEN, align: "center" }) ],
|
| 732 |
+
[ fCell("Expression \u2192 GRN"), fCell("Predict first, understand later"), fCell("Suboptimal", { color: ORANGE, align: "center" }) ],
|
| 733 |
+
[ fCell("Simultaneous", fAlt), fCell("No explicit order", fAlt), fCell("Worst", { ...fAlt, color: RED, align: "center" }) ]
|
| 734 |
+
], {
|
| 735 |
+
x: 0.6, y: 3.15, w: 8.8,
|
| 736 |
+
colW: [2.5, 3.8, 2.5],
|
| 737 |
+
border: { pt: 0.5, color: "DDE1E6" },
|
| 738 |
+
rowH: [0.4, 0.4, 0.4, 0.4]
|
| 739 |
+
});
|
| 740 |
+
|
| 741 |
+
s13.addText("If \"GRN \u2192 Expression\" wins: GRN understanding is a prerequisite, not a byproduct.", {
|
| 742 |
+
x: 0.6, y: 4.8, w: 8.8, h: 0.4,
|
| 743 |
+
fontSize: 12, fontFace: BF, color: NAVY, bold: true, italic: true
|
| 744 |
+
});
|
| 745 |
+
|
| 746 |
+
// ════════════════════════════════════════════════════════════
|
| 747 |
+
// SLIDE 14: TAKE-HOME MESSAGE
|
| 748 |
+
// ════════════════════════════════════════════════════════════
|
| 749 |
+
let s14 = pres.addSlide();
|
| 750 |
+
s14.background = { color: NAVY };
|
| 751 |
+
s14.addShape(pres.shapes.RECTANGLE, {
|
| 752 |
+
x: 0, y: 0, w: 10, h: 0.06, fill: { color: BLUE }
|
| 753 |
+
});
|
| 754 |
+
s14.addText("Take-Home Message", {
|
| 755 |
+
x: 1, y: 1.0, w: 8, h: 0.6,
|
| 756 |
+
fontSize: 24, fontFace: HF, color: LT_BLUE, align: "center"
|
| 757 |
+
});
|
| 758 |
+
s14.addText([
|
| 759 |
+
{ text: "We embed biological prior \u2014 perturbation cascades through GRN \u2014", options: { breakLine: true } },
|
| 760 |
+
{ text: "into generative modeling via cascaded flow matching,", options: { breakLine: true } },
|
| 761 |
+
{ text: "forcing the model to ", options: {} },
|
| 762 |
+
{ text: "\"understand regulatory rewiring", options: { bold: true } },
|
| 763 |
+
{ text: "", options: { breakLine: true } },
|
| 764 |
+
{ text: "before predicting expression changes.\"", options: { bold: true } }
|
| 765 |
+
], {
|
| 766 |
+
x: 1.0, y: 2.0, w: 8.0, h: 2.0,
|
| 767 |
+
fontSize: 18, fontFace: BF, color: WHITE, align: "center", valign: "middle",
|
| 768 |
+
paraSpaceAfter: 6
|
| 769 |
+
});
|
| 770 |
+
s14.addShape(pres.shapes.RECTANGLE, {
|
| 771 |
+
x: 3.8, y: 4.3, w: 2.4, h: 0.035, fill: { color: BLUE }
|
| 772 |
+
});
|
| 773 |
+
s14.addText("Thank You", {
|
| 774 |
+
x: 1, y: 4.5, w: 8, h: 0.5,
|
| 775 |
+
fontSize: 20, fontFace: HF, color: TXT_LT, align: "center"
|
| 776 |
+
});
|
| 777 |
+
|
| 778 |
+
// ─── WRITE ───
|
| 779 |
+
pres.writeFile({ fileName: "/home/hp250092/ku50001222/qian/aivc/lfj/Report/PPT2/GRN_CCFM_presentation.pptx" })
|
| 780 |
+
.then(() => console.log("SUCCESS: Presentation saved."))
|
| 781 |
+
.catch(err => console.error("ERROR:", err));
|
Report/PPT2/slide-01.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-02.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-03.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-04.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-05.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-06.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-07.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-08.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-09.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-10.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-11.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-12.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-13.jpg
ADDED
|
Git LFS Details
|
Report/PPT2/slide-14.jpg
ADDED
|
Git LFS Details
|
Report/build_pptx.js
ADDED
|
@@ -0,0 +1,703 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
const pptxgen = require("pptxgenjs");
|
| 2 |
+
|
| 3 |
+
const pres = new pptxgen();
|
| 4 |
+
pres.layout = "LAYOUT_16x9";
|
| 5 |
+
pres.author = "Qian";
|
| 6 |
+
pres.title = "GRN-Guided Cascaded Flow Matching";
|
| 7 |
+
|
| 8 |
+
// === Color Palette ===
|
| 9 |
+
const C = {
|
| 10 |
+
navy: "0B1D3A",
|
| 11 |
+
deepBlue: "0E3B5C",
|
| 12 |
+
teal: "0D7377",
|
| 13 |
+
seafoam: "14B8A6",
|
| 14 |
+
mint: "99F6E4", // brightened for dark bg readability
|
| 15 |
+
gold: "F59E0B",
|
| 16 |
+
orange: "F97316",
|
| 17 |
+
coral: "EF4444",
|
| 18 |
+
white: "FFFFFF",
|
| 19 |
+
offWhite: "F0F4F8",
|
| 20 |
+
lightGray: "E2E8F0",
|
| 21 |
+
midGray: "94A3B8",
|
| 22 |
+
darkGray: "334155",
|
| 23 |
+
textDark: "1E293B",
|
| 24 |
+
textMid: "475569",
|
| 25 |
+
accent1: "3B82F6", // blue for expression
|
| 26 |
+
accent2: "F59E0B", // gold for GRN/latent
|
| 27 |
+
accent3: "10B981", // green for bio
|
| 28 |
+
subtitleOnDark: "A7F3D0", // bright mint-green for subtitles on navy
|
| 29 |
+
};
|
| 30 |
+
|
| 31 |
+
const cardShadow = () => ({ type: "outer", blur: 4, offset: 2, angle: 135, color: "000000", opacity: 0.10 });
|
| 32 |
+
|
| 33 |
+
// Slide number — placed safely out of content area
|
| 34 |
+
function addSlideNum(slide, num) {
|
| 35 |
+
slide.addText(String(num), {
|
| 36 |
+
x: 9.3, y: 5.2, w: 0.5, h: 0.3,
|
| 37 |
+
fontSize: 8, color: C.midGray, align: "right", fontFace: "Calibri",
|
| 38 |
+
});
|
| 39 |
+
}
|
| 40 |
+
|
| 41 |
+
// Section divider — centered vertically, improved contrast
|
| 42 |
+
function addDividerSlide(title, subtitle, num) {
|
| 43 |
+
const s = pres.addSlide();
|
| 44 |
+
s.background = { color: C.navy };
|
| 45 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.7, y: 2.0, w: 0.8, h: 0.06, fill: { color: C.seafoam } });
|
| 46 |
+
s.addText(title, {
|
| 47 |
+
x: 0.7, y: 2.2, w: 8.6, h: 1.0,
|
| 48 |
+
fontSize: 36, fontFace: "Georgia", color: C.white, bold: true, margin: 0,
|
| 49 |
+
});
|
| 50 |
+
if (subtitle) {
|
| 51 |
+
s.addText(subtitle, {
|
| 52 |
+
x: 0.7, y: 3.3, w: 8.6, h: 0.6,
|
| 53 |
+
fontSize: 16, fontFace: "Calibri", color: C.subtitleOnDark, margin: 0,
|
| 54 |
+
});
|
| 55 |
+
}
|
| 56 |
+
addSlideNum(s, num);
|
| 57 |
+
return s;
|
| 58 |
+
}
|
| 59 |
+
|
| 60 |
+
// Content slide — title with teal top bar
|
| 61 |
+
function addContentSlide(title, num) {
|
| 62 |
+
const s = pres.addSlide();
|
| 63 |
+
s.background = { color: C.offWhite };
|
| 64 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0, y: 0, w: 10, h: 0.06, fill: { color: C.teal } });
|
| 65 |
+
s.addText(title, {
|
| 66 |
+
x: 0.6, y: 0.15, w: 8.8, h: 0.55,
|
| 67 |
+
fontSize: 22, fontFace: "Georgia", color: C.textDark, bold: true, margin: 0,
|
| 68 |
+
});
|
| 69 |
+
addSlideNum(s, num);
|
| 70 |
+
return s;
|
| 71 |
+
}
|
| 72 |
+
|
| 73 |
+
let slideNum = 0;
|
| 74 |
+
|
| 75 |
+
// ============================================================
|
| 76 |
+
// SLIDE 1: Title
|
| 77 |
+
// ============================================================
|
| 78 |
+
slideNum++;
|
| 79 |
+
{
|
| 80 |
+
const s = pres.addSlide();
|
| 81 |
+
s.background = { color: C.navy };
|
| 82 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0, y: 0, w: 10, h: 0.08, fill: { color: C.seafoam } });
|
| 83 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.7, y: 1.2, w: 1.2, h: 0.06, fill: { color: C.gold } });
|
| 84 |
+
|
| 85 |
+
s.addText("GRN-Guided\nCascaded Flow Matching\nfor Single-Cell Perturbation Prediction", {
|
| 86 |
+
x: 0.7, y: 1.4, w: 8.6, h: 2.2,
|
| 87 |
+
fontSize: 30, fontFace: "Georgia", color: C.white, bold: true, margin: 0,
|
| 88 |
+
lineSpacingMultiple: 1.35,
|
| 89 |
+
});
|
| 90 |
+
|
| 91 |
+
s.addText("Gene Regulatory Network meets Flow Matching", {
|
| 92 |
+
x: 0.7, y: 3.75, w: 8.6, h: 0.4,
|
| 93 |
+
fontSize: 14, fontFace: "Calibri", color: C.subtitleOnDark, italic: true, margin: 0,
|
| 94 |
+
});
|
| 95 |
+
|
| 96 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.7, y: 4.35, w: 2.5, h: 0.02, fill: { color: C.midGray } });
|
| 97 |
+
|
| 98 |
+
s.addText("Group Meeting | 2026.03", {
|
| 99 |
+
x: 0.7, y: 4.5, w: 8.6, h: 0.4,
|
| 100 |
+
fontSize: 12, fontFace: "Calibri", color: C.lightGray, margin: 0,
|
| 101 |
+
});
|
| 102 |
+
addSlideNum(s, slideNum);
|
| 103 |
+
}
|
| 104 |
+
|
| 105 |
+
// ============================================================
|
| 106 |
+
// SLIDE 2: Section — Task
|
| 107 |
+
// ============================================================
|
| 108 |
+
slideNum++;
|
| 109 |
+
addDividerSlide("1. Task", "Single-Cell Perturbation Prediction", slideNum);
|
| 110 |
+
|
| 111 |
+
// ============================================================
|
| 112 |
+
// SLIDE 3: Virtual Cell + Perturbation Types
|
| 113 |
+
// ============================================================
|
| 114 |
+
slideNum++;
|
| 115 |
+
{
|
| 116 |
+
const s = addContentSlide("Virtual Cell & Perturbation Types", slideNum);
|
| 117 |
+
|
| 118 |
+
// Virtual Cell callout
|
| 119 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 4.2, h: 1.15, fill: { color: C.white }, shadow: cardShadow() });
|
| 120 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 0.07, h: 1.15, fill: { color: C.teal } });
|
| 121 |
+
s.addText([
|
| 122 |
+
{ text: "Virtual Cell", options: { bold: true, fontSize: 13, color: C.teal, breakLine: true } },
|
| 123 |
+
{ text: "AI model simulating real cell behavior: given genotype, environment, perturbation \u2192 predict molecular state changes. Perturbation prediction is its most critical subtask.", options: { fontSize: 10.5, color: C.textMid } },
|
| 124 |
+
], { x: 0.75, y: 0.95, w: 3.8, h: 1.05, valign: "top", fontFace: "Calibri", margin: 0 });
|
| 125 |
+
|
| 126 |
+
// Three perturbation type cards (right)
|
| 127 |
+
const types = [
|
| 128 |
+
{ title: "Drug Perturbation", desc: "Small molecules / drugs (L1000/LINCS)", color: C.accent1 },
|
| 129 |
+
{ title: "Cytokine Perturbation", desc: "Cytokines (IL-6, TNF-a, IFN-g) signaling", color: C.accent3 },
|
| 130 |
+
{ title: "Genetic Perturbation", desc: "CRISPR KO / CRISPRa OE / RNAi KD", color: C.accent2 },
|
| 131 |
+
];
|
| 132 |
+
const cardX = 5.0, cardW = 4.5, cardH = 0.7;
|
| 133 |
+
types.forEach((t, i) => {
|
| 134 |
+
const yy = 0.9 + i * (cardH + 0.12);
|
| 135 |
+
s.addShape(pres.shapes.RECTANGLE, { x: cardX, y: yy, w: cardW, h: cardH, fill: { color: C.white }, shadow: cardShadow() });
|
| 136 |
+
s.addShape(pres.shapes.RECTANGLE, { x: cardX, y: yy, w: 0.07, h: cardH, fill: { color: t.color } });
|
| 137 |
+
s.addText(t.title, {
|
| 138 |
+
x: cardX + 0.2, y: yy + 0.05, w: 4.0, h: 0.28,
|
| 139 |
+
fontSize: 11.5, fontFace: "Calibri", bold: true, color: C.textDark, margin: 0,
|
| 140 |
+
});
|
| 141 |
+
s.addText(t.desc, {
|
| 142 |
+
x: cardX + 0.2, y: yy + 0.35, w: 4.0, h: 0.3,
|
| 143 |
+
fontSize: 9.5, fontFace: "Calibri", color: C.textMid, margin: 0,
|
| 144 |
+
});
|
| 145 |
+
});
|
| 146 |
+
|
| 147 |
+
// Focus banner
|
| 148 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 3.4, w: 9.0, h: 0.5, fill: { color: C.navy } });
|
| 149 |
+
s.addText("This work: genetic perturbation (Perturb-seq) = CRISPR perturbation + scRNA-seq readout", {
|
| 150 |
+
x: 0.7, y: 3.42, w: 8.6, h: 0.46,
|
| 151 |
+
fontSize: 11.5, fontFace: "Calibri", color: C.white, bold: true, margin: 0, valign: "middle",
|
| 152 |
+
});
|
| 153 |
+
|
| 154 |
+
// Task formalization
|
| 155 |
+
s.addText([
|
| 156 |
+
{ text: "Task: ", options: { bold: true, color: C.teal, fontSize: 13 } },
|
| 157 |
+
{ text: "x_ctrl + perturbation ID \u2192 predict x_pert (x \u2208 R^G, G \u2248 5000 HVG)", options: { color: C.textDark, fontSize: 12, fontFace: "Consolas" } },
|
| 158 |
+
], { x: 0.5, y: 4.05, w: 9.0, h: 0.35, fontFace: "Calibri", margin: 0 });
|
| 159 |
+
|
| 160 |
+
// Key challenges
|
| 161 |
+
s.addText([
|
| 162 |
+
{ text: "Drug screening acceleration | Combinatorial explosion: N genes \u2192 N(N-1)/2 combos | ", options: { fontSize: 10, color: C.textMid, breakLine: false } },
|
| 163 |
+
{ text: "No paired data (destructive measurement)", options: { fontSize: 10, color: C.coral, bold: true } },
|
| 164 |
+
], { x: 0.5, y: 4.45, w: 9.0, h: 0.35, fontFace: "Calibri", margin: 0 });
|
| 165 |
+
}
|
| 166 |
+
|
| 167 |
+
// ============================================================
|
| 168 |
+
// SLIDE 4: Section — Existing Methods
|
| 169 |
+
// ============================================================
|
| 170 |
+
slideNum++;
|
| 171 |
+
addDividerSlide("2. Existing Methods", "And their common blind spot", slideNum);
|
| 172 |
+
|
| 173 |
+
// ============================================================
|
| 174 |
+
// SLIDE 5: Methods Overview Table
|
| 175 |
+
// ============================================================
|
| 176 |
+
slideNum++;
|
| 177 |
+
{
|
| 178 |
+
const s = addContentSlide("Existing Methods: Overview", slideNum);
|
| 179 |
+
|
| 180 |
+
const methods = [
|
| 181 |
+
{ name: "Additive Shift", cat: "Baseline", approach: "Mean shift: x = x_ctrl + delta_mean", issue: "Ignores cell heterogeneity" },
|
| 182 |
+
{ name: "scGPT", cat: "Foundation Model", approach: "Masked token completion (fine-tune)", issue: "Encodes absolute state, not change" },
|
| 183 |
+
{ name: "Geneformer", cat: "Foundation Model", approach: "In-silico: delete gene token", issue: "Heuristic, no learned dynamics" },
|
| 184 |
+
{ name: "CPA", cat: "Dedicated Model", approach: "VAE: basal + perturbation (additive)", issue: "Linear additivity too strong" },
|
| 185 |
+
{ name: "GEARS", cat: "Dedicated Model", approach: "GNN on GO graph + cross-attention", issue: "Static prior graph, deterministic" },
|
| 186 |
+
{ name: "STATE", cat: "Dedicated Model", approach: "Stacked attention on expression", issue: "Deterministic, no GRN modeling" },
|
| 187 |
+
{ name: "CellFlow", cat: "Flow Matching", approach: "FM + pretrained embedding cond.", issue: "Embedding = absolute state" },
|
| 188 |
+
{ name: "scDFM", cat: "Flow Matching", approach: "Conditional FM + DiffPerceiver", issue: "No GRN understanding" },
|
| 189 |
+
];
|
| 190 |
+
|
| 191 |
+
const hY = 0.85;
|
| 192 |
+
const cols = [
|
| 193 |
+
{ x: 0.5, w: 1.5, label: "Method" },
|
| 194 |
+
{ x: 2.0, w: 1.5, label: "Category" },
|
| 195 |
+
{ x: 3.5, w: 3.2, label: "Approach" },
|
| 196 |
+
{ x: 6.7, w: 2.8, label: "Key Limitation" },
|
| 197 |
+
];
|
| 198 |
+
|
| 199 |
+
// Header
|
| 200 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: hY, w: 9.0, h: 0.35, fill: { color: C.teal } });
|
| 201 |
+
cols.forEach(c => {
|
| 202 |
+
s.addText(c.label, {
|
| 203 |
+
x: c.x + 0.08, y: hY, w: c.w - 0.08, h: 0.35,
|
| 204 |
+
fontSize: 10, fontFace: "Calibri", bold: true, color: C.white, valign: "middle", margin: 0,
|
| 205 |
+
});
|
| 206 |
+
});
|
| 207 |
+
|
| 208 |
+
// Data rows
|
| 209 |
+
const rowH = 0.37;
|
| 210 |
+
methods.forEach((m, i) => {
|
| 211 |
+
const ry = hY + 0.35 + i * rowH;
|
| 212 |
+
const bgColor = i % 2 === 0 ? C.white : "F8FAFC";
|
| 213 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: ry, w: 9.0, h: rowH, fill: { color: bgColor } });
|
| 214 |
+
s.addText(m.name, {
|
| 215 |
+
x: cols[0].x + 0.08, y: ry, w: cols[0].w - 0.08, h: rowH,
|
| 216 |
+
fontSize: 9.5, fontFace: "Calibri", bold: true, color: C.textDark, valign: "middle", margin: 0,
|
| 217 |
+
});
|
| 218 |
+
s.addText(m.cat, {
|
| 219 |
+
x: cols[1].x + 0.08, y: ry, w: cols[1].w - 0.08, h: rowH,
|
| 220 |
+
fontSize: 9, fontFace: "Calibri", color: C.textMid, valign: "middle", margin: 0,
|
| 221 |
+
});
|
| 222 |
+
s.addText(m.approach, {
|
| 223 |
+
x: cols[2].x + 0.08, y: ry, w: cols[2].w - 0.08, h: rowH,
|
| 224 |
+
fontSize: 9, fontFace: "Calibri", color: C.textDark, valign: "middle", margin: 0,
|
| 225 |
+
});
|
| 226 |
+
s.addText(m.issue, {
|
| 227 |
+
x: cols[3].x + 0.08, y: ry, w: cols[3].w - 0.08, h: rowH,
|
| 228 |
+
fontSize: 9, fontFace: "Calibri", color: C.coral, bold: true, valign: "middle", margin: 0,
|
| 229 |
+
});
|
| 230 |
+
});
|
| 231 |
+
|
| 232 |
+
// Common blind spot callout
|
| 233 |
+
const bY = hY + 0.35 + methods.length * rowH + 0.3;
|
| 234 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: bY, w: 9.0, h: 0.7, fill: { color: C.navy } });
|
| 235 |
+
s.addText([
|
| 236 |
+
{ text: "Common blind spot: ", options: { bold: true, color: C.gold, fontSize: 13 } },
|
| 237 |
+
{ text: "Perturbation \u2192 [black box] \u2192 Expression change", options: { color: C.white, fontSize: 13, breakLine: true } },
|
| 238 |
+
{ text: "No method explicitly models: Perturbation \u2192 GRN rewiring \u2192 Expression change", options: { color: C.subtitleOnDark, fontSize: 11 } },
|
| 239 |
+
], { x: 0.7, y: bY + 0.03, w: 8.6, h: 0.65, fontFace: "Calibri", valign: "middle", margin: 0 });
|
| 240 |
+
}
|
| 241 |
+
|
| 242 |
+
// ============================================================
|
| 243 |
+
// SLIDE 6: Section — Motivation
|
| 244 |
+
// ============================================================
|
| 245 |
+
slideNum++;
|
| 246 |
+
addDividerSlide("3. Motivation", "Why GRN + Flow Matching?", slideNum);
|
| 247 |
+
|
| 248 |
+
// ============================================================
|
| 249 |
+
// SLIDE 7: Motivation 1 — Flow Matching
|
| 250 |
+
// ============================================================
|
| 251 |
+
slideNum++;
|
| 252 |
+
{
|
| 253 |
+
const s = addContentSlide("Motivation 1: Flow Matching for Unpaired Data", slideNum);
|
| 254 |
+
|
| 255 |
+
// Problem card (left)
|
| 256 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 4.2, h: 1.8, fill: { color: C.white }, shadow: cardShadow() });
|
| 257 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 0.07, h: 1.8, fill: { color: C.coral } });
|
| 258 |
+
s.addText([
|
| 259 |
+
{ text: "The Pairing Problem", options: { bold: true, fontSize: 13, color: C.coral, breakLine: true } },
|
| 260 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 261 |
+
{ text: "Perturbation is destructive:", options: { fontSize: 11, color: C.textDark, breakLine: true } },
|
| 262 |
+
{ text: "One cell measured ONCE only", options: { fontSize: 11, color: C.textDark, breakLine: true } },
|
| 263 |
+
{ text: "No (x_ctrl, x_pert) pairs available", options: { fontSize: 11, color: C.coral, bold: true, breakLine: true } },
|
| 264 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 265 |
+
{ text: "Mean matching \u2192 loses heterogeneity", options: { bullet: true, fontSize: 10, color: C.textMid, breakLine: true } },
|
| 266 |
+
{ text: "Autoencoder \u2192 limited reconstruction", options: { bullet: true, fontSize: 10, color: C.textMid } },
|
| 267 |
+
], { x: 0.75, y: 0.95, w: 3.8, h: 1.7, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 268 |
+
|
| 269 |
+
// Solution card (right)
|
| 270 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.0, y: 0.9, w: 4.5, h: 1.8, fill: { color: C.white }, shadow: cardShadow() });
|
| 271 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.0, y: 0.9, w: 0.07, h: 1.8, fill: { color: C.accent3 } });
|
| 272 |
+
s.addText([
|
| 273 |
+
{ text: "Flow Matching Solution", options: { bold: true, fontSize: 13, color: C.accent3, breakLine: true } },
|
| 274 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 275 |
+
{ text: "Learn probabilistic transport mapping\nbetween distributions (not individual cells)", options: { fontSize: 11, color: C.textDark, breakLine: true } },
|
| 276 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 277 |
+
{ text: "Only needs population-level distributions", options: { bullet: true, fontSize: 10, color: C.textMid, breakLine: true } },
|
| 278 |
+
{ text: "Conditional OT for efficient pairing", options: { bullet: true, fontSize: 10, color: C.textMid, breakLine: true } },
|
| 279 |
+
{ text: "Generative output = uncertainty estimation", options: { bullet: true, fontSize: 10, color: C.textMid } },
|
| 280 |
+
], { x: 5.25, y: 0.95, w: 4.1, h: 1.7, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 281 |
+
|
| 282 |
+
// Flow diagram
|
| 283 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 3.0, w: 9.0, h: 1.6, fill: { color: C.white }, shadow: cardShadow() });
|
| 284 |
+
|
| 285 |
+
s.addShape(pres.shapes.OVAL, { x: 1.2, y: 3.25, w: 1.8, h: 1.1, fill: { color: "FEE2E2" } });
|
| 286 |
+
s.addText("noise x\u2080", { x: 1.2, y: 3.25, w: 1.8, h: 1.1, fontSize: 12, fontFace: "Calibri", color: C.coral, align: "center", valign: "middle", bold: true, margin: 0 });
|
| 287 |
+
|
| 288 |
+
s.addText("v\u03B8( x, t, ctrl, pert )", {
|
| 289 |
+
x: 3.2, y: 3.45, w: 3.6, h: 0.5,
|
| 290 |
+
fontSize: 14, fontFace: "Consolas", color: C.teal, align: "center", valign: "middle", bold: true, margin: 0,
|
| 291 |
+
});
|
| 292 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 3.5, y: 3.95, w: 3.0, h: 0.04, fill: { color: C.teal } });
|
| 293 |
+
s.addText("learned velocity field (ODE)", {
|
| 294 |
+
x: 3.2, y: 4.0, w: 3.6, h: 0.3,
|
| 295 |
+
fontSize: 9, fontFace: "Calibri", color: C.textMid, align: "center", margin: 0,
|
| 296 |
+
});
|
| 297 |
+
|
| 298 |
+
s.addShape(pres.shapes.OVAL, { x: 7.0, y: 3.25, w: 1.8, h: 1.1, fill: { color: "D1FAE5" } });
|
| 299 |
+
s.addText("predicted\nx_pert", { x: 7.0, y: 3.25, w: 1.8, h: 1.1, fontSize: 12, fontFace: "Calibri", color: C.accent3, align: "center", valign: "middle", bold: true, margin: 0 });
|
| 300 |
+
}
|
| 301 |
+
|
| 302 |
+
// ============================================================
|
| 303 |
+
// SLIDE 8: Motivation 2 — GRN Cascade
|
| 304 |
+
// ============================================================
|
| 305 |
+
slideNum++;
|
| 306 |
+
{
|
| 307 |
+
const s = addContentSlide("Motivation 2: Perturbation Propagates via GRN", slideNum);
|
| 308 |
+
|
| 309 |
+
// Cascade diagram (left)
|
| 310 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 5.5, h: 2.8, fill: { color: C.white }, shadow: cardShadow() });
|
| 311 |
+
|
| 312 |
+
const steps = [
|
| 313 |
+
{ text: "CRISPR knock-out Gene A", color: C.coral, bold: true },
|
| 314 |
+
{ text: "Gene A expression --> 0", color: C.coral, bold: false },
|
| 315 |
+
{ text: "Direct targets B, C, D change (1st order)", color: C.accent2, bold: false },
|
| 316 |
+
{ text: "B->E,F C->G,H D->I ... (cascade)", color: C.accent2, bold: false },
|
| 317 |
+
{ text: "Thousands of genes ultimately affected", color: C.teal, bold: true },
|
| 318 |
+
];
|
| 319 |
+
steps.forEach((st, i) => {
|
| 320 |
+
const yy = 1.05 + i * 0.45;
|
| 321 |
+
s.addText((i > 0 ? " | " : " ") + st.text, {
|
| 322 |
+
x: 0.8, y: yy, w: 5.0, h: 0.38,
|
| 323 |
+
fontSize: 11, fontFace: "Calibri", color: st.color, bold: st.bold, margin: 0,
|
| 324 |
+
});
|
| 325 |
+
});
|
| 326 |
+
|
| 327 |
+
s.addText("This cascade path = Gene Regulatory Network (GRN)", {
|
| 328 |
+
x: 0.8, y: 3.3, w: 5.0, h: 0.3,
|
| 329 |
+
fontSize: 11, fontFace: "Calibri", color: C.navy, bold: true, italic: true, margin: 0,
|
| 330 |
+
});
|
| 331 |
+
|
| 332 |
+
// Comparison cards (right)
|
| 333 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 6.3, y: 0.9, w: 3.2, h: 1.2, fill: { color: "FEF3C7" }, shadow: cardShadow() });
|
| 334 |
+
s.addText([
|
| 335 |
+
{ text: "Existing Methods", options: { bold: true, fontSize: 12, color: C.textDark, breakLine: true } },
|
| 336 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 337 |
+
{ text: "Pert -> [black box] -> Expr", options: { fontSize: 11, fontFace: "Consolas", color: C.coral, breakLine: true } },
|
| 338 |
+
{ text: "End-to-end, no GRN understanding", options: { fontSize: 10, color: C.textMid } },
|
| 339 |
+
], { x: 6.5, y: 0.95, w: 2.9, h: 1.1, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 340 |
+
|
| 341 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 6.3, y: 2.3, w: 3.2, h: 1.4, fill: { color: "D1FAE5" }, shadow: cardShadow() });
|
| 342 |
+
s.addText([
|
| 343 |
+
{ text: "Our Approach", options: { bold: true, fontSize: 12, color: C.textDark, breakLine: true } },
|
| 344 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 345 |
+
{ text: "Pert -> GRN change -> Expr", options: { fontSize: 11, fontFace: "Consolas", color: C.accent3, breakLine: true } },
|
| 346 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 347 |
+
{ text: "Explicitly model how perturbation rewires the regulatory network, then predict expression", options: { fontSize: 10, color: C.textDark } },
|
| 348 |
+
], { x: 6.5, y: 2.35, w: 2.9, h: 1.3, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 349 |
+
|
| 350 |
+
// Bottom insight
|
| 351 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 4.1, w: 9.0, h: 0.5, fill: { color: C.navy } });
|
| 352 |
+
s.addText("Understanding GRN changes is a prerequisite for accurate expression prediction", {
|
| 353 |
+
x: 0.7, y: 4.12, w: 8.6, h: 0.46,
|
| 354 |
+
fontSize: 12, fontFace: "Calibri", color: C.gold, bold: true, margin: 0, valign: "middle",
|
| 355 |
+
});
|
| 356 |
+
}
|
| 357 |
+
|
| 358 |
+
// ============================================================
|
| 359 |
+
// SLIDE 9: Motivation 3 — scGPT Attention = GRN
|
| 360 |
+
// ============================================================
|
| 361 |
+
slideNum++;
|
| 362 |
+
{
|
| 363 |
+
const s = addContentSlide("Motivation 3: scGPT Attention = Data-Driven GRN", slideNum);
|
| 364 |
+
|
| 365 |
+
// Left: explanation
|
| 366 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 4.5, h: 2.1, fill: { color: C.white }, shadow: cardShadow() });
|
| 367 |
+
s.addText([
|
| 368 |
+
{ text: "scGPT Transformer Attention", options: { bold: true, fontSize: 13, color: C.teal, breakLine: true } },
|
| 369 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 370 |
+
{ text: "attn[i][j] high -> gene j influences gene i", options: { fontSize: 11, fontFace: "Consolas", color: C.textDark, breakLine: true } },
|
| 371 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 372 |
+
{ text: "= Context-dependent, data-driven GRN", options: { fontSize: 12, color: C.navy, bold: true, breakLine: true } },
|
| 373 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 374 |
+
{ text: "vs static GO graph:", options: { bold: true, fontSize: 10, color: C.textMid, breakLine: true } },
|
| 375 |
+
{ text: "Changes with cell state (context-aware)", options: { bullet: true, fontSize: 10, color: C.textMid, breakLine: true } },
|
| 376 |
+
{ text: "Learned from massive scRNA-seq data", options: { bullet: true, fontSize: 10, color: C.textMid, breakLine: true } },
|
| 377 |
+
{ text: "Captures non-linear regulatory logic", options: { bullet: true, fontSize: 10, color: C.textMid } },
|
| 378 |
+
], { x: 0.7, y: 0.95, w: 4.1, h: 2.0, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 379 |
+
|
| 380 |
+
// Right: Attention-Delta
|
| 381 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.3, y: 0.9, w: 4.2, h: 2.1, fill: { color: C.white }, shadow: cardShadow() });
|
| 382 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.3, y: 0.9, w: 0.07, h: 2.1, fill: { color: C.gold } });
|
| 383 |
+
s.addText([
|
| 384 |
+
{ text: "Attention-Delta", options: { bold: true, fontSize: 13, color: C.accent2, breakLine: true } },
|
| 385 |
+
{ text: "", options: { breakLine: true, fontSize: 5 } },
|
| 386 |
+
{ text: "Same frozen scGPT, two inputs:", options: { fontSize: 11, color: C.textDark, breakLine: true } },
|
| 387 |
+
{ text: "attn_ctrl = scGPT(x_ctrl)", options: { fontSize: 10.5, fontFace: "Consolas", color: C.accent1, breakLine: true } },
|
| 388 |
+
{ text: "attn_pert = scGPT(x_pert)", options: { fontSize: 10.5, fontFace: "Consolas", color: C.coral, breakLine: true } },
|
| 389 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 390 |
+
{ text: "delta_attn = attn_pert - attn_ctrl", options: { fontSize: 11, fontFace: "Consolas", color: C.navy, bold: true, breakLine: true } },
|
| 391 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 392 |
+
{ text: "Directly captures how perturbation\nrewires gene regulatory relationships", options: { fontSize: 10, color: C.textDark } },
|
| 393 |
+
], { x: 5.55, y: 0.95, w: 3.8, h: 2.0, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 394 |
+
|
| 395 |
+
// Bottom: GRN features formula
|
| 396 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 3.3, w: 9.0, h: 1.25, fill: { color: C.navy } });
|
| 397 |
+
s.addText([
|
| 398 |
+
{ text: "GRN Change Features:", options: { bold: true, fontSize: 14, color: C.gold, breakLine: true } },
|
| 399 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 400 |
+
{ text: "z = delta_attn x gene_embeddings", options: { fontSize: 17, fontFace: "Consolas", color: C.white, breakLine: true } },
|
| 401 |
+
{ text: " (G x G) (G x 512) --> (G x 512)", options: { fontSize: 11, fontFace: "Consolas", color: C.subtitleOnDark } },
|
| 402 |
+
], { x: 0.7, y: 3.35, w: 8.6, h: 1.15, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 403 |
+
}
|
| 404 |
+
|
| 405 |
+
// ============================================================
|
| 406 |
+
// SLIDE 10: Section — Our Method
|
| 407 |
+
// ============================================================
|
| 408 |
+
slideNum++;
|
| 409 |
+
addDividerSlide("4. Our Method", "GRN-Guided Cascaded Flow Matching", slideNum);
|
| 410 |
+
|
| 411 |
+
// ============================================================
|
| 412 |
+
// SLIDE 11: Two-Stage Cascaded FM
|
| 413 |
+
// ============================================================
|
| 414 |
+
slideNum++;
|
| 415 |
+
{
|
| 416 |
+
const s = addContentSlide("Two-Stage Cascaded Flow Matching", slideNum);
|
| 417 |
+
|
| 418 |
+
// Stage 1 card
|
| 419 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 4.0, h: 2.0, fill: { color: C.white }, shadow: cardShadow() });
|
| 420 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 0.07, h: 2.0, fill: { color: C.gold } });
|
| 421 |
+
s.addText([
|
| 422 |
+
{ text: "Stage 1: GRN Latent Flow", options: { bold: true, fontSize: 13, color: C.accent2, breakLine: true } },
|
| 423 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 424 |
+
{ text: "noise ==(ODE)==> GRN features", options: { fontSize: 12, fontFace: "Consolas", color: C.textDark, breakLine: true } },
|
| 425 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 426 |
+
{ text: "\"Understand how gene regulation\n changes under perturbation\"", options: { fontSize: 11, color: C.accent2, italic: true, breakLine: true } },
|
| 427 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 428 |
+
{ text: "t_latent: 0 -> 1", options: { fontSize: 10, fontFace: "Consolas", color: C.textMid, breakLine: true } },
|
| 429 |
+
{ text: "t_expr = 0 (expression frozen)", options: { fontSize: 10, fontFace: "Consolas", color: C.textMid } },
|
| 430 |
+
], { x: 0.75, y: 0.95, w: 3.6, h: 1.9, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 431 |
+
|
| 432 |
+
// Arrow
|
| 433 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 4.6, y: 1.75, w: 0.7, h: 0.04, fill: { color: C.teal } });
|
| 434 |
+
s.addText(">", { x: 5.0, y: 1.55, w: 0.5, h: 0.5, fontSize: 24, color: C.teal, align: "center", valign: "middle", fontFace: "Calibri", bold: true, margin: 0 });
|
| 435 |
+
|
| 436 |
+
// Stage 2 card
|
| 437 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.5, y: 0.9, w: 4.0, h: 2.0, fill: { color: C.white }, shadow: cardShadow() });
|
| 438 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.5, y: 0.9, w: 0.07, h: 2.0, fill: { color: C.accent1 } });
|
| 439 |
+
s.addText([
|
| 440 |
+
{ text: "Stage 2: Expression Flow", options: { bold: true, fontSize: 13, color: C.accent1, breakLine: true } },
|
| 441 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 442 |
+
{ text: "noise ==(ODE)==> expression", options: { fontSize: 12, fontFace: "Consolas", color: C.textDark, breakLine: true } },
|
| 443 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 444 |
+
{ text: "\"Based on GRN understanding,\n predict gene expression changes\"", options: { fontSize: 11, color: C.accent1, italic: true, breakLine: true } },
|
| 445 |
+
{ text: "", options: { breakLine: true, fontSize: 6 } },
|
| 446 |
+
{ text: "t_expr: 0 -> 1", options: { fontSize: 10, fontFace: "Consolas", color: C.textMid, breakLine: true } },
|
| 447 |
+
{ text: "t_latent = 1 (GRN complete)", options: { fontSize: 10, fontFace: "Consolas", color: C.textMid } },
|
| 448 |
+
], { x: 5.75, y: 0.95, w: 3.6, h: 1.9, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 449 |
+
|
| 450 |
+
// Bio intuition banner
|
| 451 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 3.15, w: 9.0, h: 0.55, fill: { color: C.navy } });
|
| 452 |
+
s.addText("Bio intuition: First understand HOW regulation changes, THEN predict WHAT expression changes", {
|
| 453 |
+
x: 0.7, y: 3.18, w: 8.6, h: 0.5,
|
| 454 |
+
fontSize: 12, fontFace: "Calibri", color: C.gold, bold: true, margin: 0, valign: "middle",
|
| 455 |
+
});
|
| 456 |
+
|
| 457 |
+
// Training note card
|
| 458 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 3.95, w: 9.0, h: 0.85, fill: { color: C.white }, shadow: cardShadow() });
|
| 459 |
+
s.addText([
|
| 460 |
+
{ text: "Cascaded Training: ", options: { bold: true, fontSize: 12, color: C.teal } },
|
| 461 |
+
{ text: "Probabilistic switching (not simultaneous)", options: { fontSize: 12, color: C.textDark, breakLine: true } },
|
| 462 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 463 |
+
{ text: "40% Train Latent Flow: t_latent random, t_expr=0, loss_latent only", options: { fontSize: 10, fontFace: "Consolas", color: C.accent2, breakLine: true } },
|
| 464 |
+
{ text: "60% Train Expr Flow: t_expr random, t_latent~1, loss_expr only", options: { fontSize: 10, fontFace: "Consolas", color: C.accent1 } },
|
| 465 |
+
], { x: 0.7, y: 4.0, w: 8.6, h: 0.75, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 466 |
+
}
|
| 467 |
+
|
| 468 |
+
// ============================================================
|
| 469 |
+
// SLIDE 12: Model Architecture
|
| 470 |
+
// ============================================================
|
| 471 |
+
slideNum++;
|
| 472 |
+
{
|
| 473 |
+
const s = addContentSlide("Model Architecture", slideNum);
|
| 474 |
+
|
| 475 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.85, w: 9.0, h: 4.2, fill: { color: C.white }, shadow: cardShadow() });
|
| 476 |
+
|
| 477 |
+
// --- Top: Two input streams ---
|
| 478 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.8, y: 1.0, w: 3.5, h: 0.6, fill: { color: "DBEAFE" } });
|
| 479 |
+
s.addText([
|
| 480 |
+
{ text: "Expression Stream", options: { bold: true, fontSize: 11, color: C.accent1, breakLine: true } },
|
| 481 |
+
{ text: "GeneEnc(id) + ValueEnc(x_t, x_ctrl) -> tokens", options: { fontSize: 8.5, fontFace: "Consolas", color: C.textMid } },
|
| 482 |
+
], { x: 0.9, y: 1.03, w: 3.3, h: 0.55, fontFace: "Calibri", valign: "middle", margin: 0 });
|
| 483 |
+
|
| 484 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.7, y: 1.0, w: 3.5, h: 0.6, fill: { color: "FEF3C7" } });
|
| 485 |
+
s.addText([
|
| 486 |
+
{ text: "Latent Stream (GRN)", options: { bold: true, fontSize: 11, color: C.accent2, breakLine: true } },
|
| 487 |
+
{ text: "LatentEmbedder(z_t) -> tokens", options: { fontSize: 8.5, fontFace: "Consolas", color: C.textMid } },
|
| 488 |
+
], { x: 5.8, y: 1.03, w: 3.3, h: 0.55, fontFace: "Calibri", valign: "middle", margin: 0 });
|
| 489 |
+
|
| 490 |
+
// Plus
|
| 491 |
+
s.addShape(pres.shapes.OVAL, { x: 4.5, y: 1.05, w: 0.5, h: 0.5, fill: { color: C.teal } });
|
| 492 |
+
s.addText("+", { x: 4.5, y: 1.05, w: 0.5, h: 0.5, fontSize: 20, color: C.white, align: "center", valign: "middle", bold: true, margin: 0 });
|
| 493 |
+
|
| 494 |
+
// Down arrow
|
| 495 |
+
s.addText("|", { x: 4.5, y: 1.6, w: 0.5, h: 0.3, fontSize: 14, color: C.teal, align: "center", valign: "middle", margin: 0 });
|
| 496 |
+
s.addText("V", { x: 4.5, y: 1.8, w: 0.5, h: 0.2, fontSize: 10, color: C.teal, align: "center", valign: "middle", margin: 0 });
|
| 497 |
+
|
| 498 |
+
// --- Shared Backbone ---
|
| 499 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 1.5, y: 2.1, w: 3.8, h: 1.5, fill: { color: C.teal } });
|
| 500 |
+
s.addText([
|
| 501 |
+
{ text: "Shared Backbone", options: { bold: true, fontSize: 13, color: C.white, breakLine: true } },
|
| 502 |
+
{ text: "", options: { breakLine: true, fontSize: 3 } },
|
| 503 |
+
{ text: "DiffPerceiverBlock x 4", options: { fontSize: 11, color: C.mint, breakLine: true } },
|
| 504 |
+
{ text: "(GeneadaLN + Adapter + DiffAttn)", options: { fontSize: 9, color: C.mint, breakLine: true } },
|
| 505 |
+
{ text: "d_model = 512", options: { fontSize: 10, fontFace: "Consolas", color: C.white } },
|
| 506 |
+
], { x: 1.6, y: 2.15, w: 3.6, h: 1.4, fontFace: "Calibri", valign: "middle", align: "center", margin: 0 });
|
| 507 |
+
|
| 508 |
+
// --- Conditioning box ---
|
| 509 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 6.0, y: 2.1, w: 3.2, h: 0.55, fill: { color: "E0E7FF" } });
|
| 510 |
+
s.addText("c = t_expr + t_latent + pert_emb", {
|
| 511 |
+
x: 6.05, y: 2.1, w: 3.1, h: 0.55,
|
| 512 |
+
fontSize: 9, fontFace: "Consolas", color: C.accent1, valign: "middle", align: "center", margin: 0,
|
| 513 |
+
});
|
| 514 |
+
s.addText("Cond.", {
|
| 515 |
+
x: 5.35, y: 2.15, w: 0.6, h: 0.45,
|
| 516 |
+
fontSize: 8, fontFace: "Calibri", color: C.textMid, valign: "middle", align: "center", margin: 0,
|
| 517 |
+
});
|
| 518 |
+
|
| 519 |
+
// --- Frozen scGPT box ---
|
| 520 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 6.0, y: 2.9, w: 3.2, h: 1.65, fill: { color: "F1F5F9" }, line: { color: C.midGray, width: 1, dashType: "dash" } });
|
| 521 |
+
s.addText([
|
| 522 |
+
{ text: "Frozen scGPT", options: { bold: true, fontSize: 11, color: C.darkGray, breakLine: true } },
|
| 523 |
+
{ text: "(no gradient)", options: { fontSize: 8, color: C.midGray, breakLine: true } },
|
| 524 |
+
{ text: "", options: { breakLine: true, fontSize: 3 } },
|
| 525 |
+
{ text: "x_ctrl, x_pert", options: { fontSize: 9, fontFace: "Consolas", color: C.accent1, breakLine: true } },
|
| 526 |
+
{ text: " -> attention layer 11", options: { fontSize: 9, fontFace: "Consolas", color: C.midGray, breakLine: true } },
|
| 527 |
+
{ text: "delta_attn x gene_emb", options: { fontSize: 9, fontFace: "Consolas", color: C.accent2, breakLine: true } },
|
| 528 |
+
{ text: " -> z_target (B,G,512)", options: { fontSize: 9, fontFace: "Consolas", color: C.accent2 } },
|
| 529 |
+
], { x: 6.1, y: 2.95, w: 3.0, h: 1.55, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 530 |
+
|
| 531 |
+
// --- Two decoder heads ---
|
| 532 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 1.5, y: 3.9, w: 1.7, h: 0.6, fill: { color: "DBEAFE" } });
|
| 533 |
+
s.addText([
|
| 534 |
+
{ text: "Expr Head", options: { bold: true, fontSize: 10, color: C.accent1, breakLine: true } },
|
| 535 |
+
{ text: "v_expr (B,G)", options: { fontSize: 9, fontFace: "Consolas", color: C.textMid } },
|
| 536 |
+
], { x: 1.5, y: 3.93, w: 1.7, h: 0.55, fontFace: "Calibri", valign: "middle", align: "center", margin: 0 });
|
| 537 |
+
|
| 538 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 3.6, y: 3.9, w: 1.7, h: 0.6, fill: { color: "FEF3C7" } });
|
| 539 |
+
s.addText([
|
| 540 |
+
{ text: "Latent Head", options: { bold: true, fontSize: 10, color: C.accent2, breakLine: true } },
|
| 541 |
+
{ text: "v_latent (B,G,512)", options: { fontSize: 9, fontFace: "Consolas", color: C.textMid } },
|
| 542 |
+
], { x: 3.6, y: 3.93, w: 1.7, h: 0.55, fontFace: "Calibri", valign: "middle", align: "center", margin: 0 });
|
| 543 |
+
}
|
| 544 |
+
|
| 545 |
+
// ============================================================
|
| 546 |
+
// SLIDE 13: Section — Challenges
|
| 547 |
+
// ============================================================
|
| 548 |
+
slideNum++;
|
| 549 |
+
addDividerSlide("5. Current Challenges", "And proposed solutions", slideNum);
|
| 550 |
+
|
| 551 |
+
// ============================================================
|
| 552 |
+
// SLIDE 14: Challenges + Solutions
|
| 553 |
+
// ============================================================
|
| 554 |
+
slideNum++;
|
| 555 |
+
{
|
| 556 |
+
const s = addContentSlide("Challenges & Solutions", slideNum);
|
| 557 |
+
|
| 558 |
+
// Challenge 1 (top-left)
|
| 559 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 4.3, h: 2.0, fill: { color: C.white }, shadow: cardShadow() });
|
| 560 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.9, w: 0.07, h: 2.0, fill: { color: C.coral } });
|
| 561 |
+
s.addText([
|
| 562 |
+
{ text: "Challenge 1: Noise in Attention", options: { bold: true, fontSize: 12, color: C.coral, breakLine: true } },
|
| 563 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 564 |
+
{ text: "Attention: 5000x5000 = 25M non-zero values", options: { fontSize: 10, color: C.textDark, breakLine: true } },
|
| 565 |
+
{ text: "Real GRN: ~20-50 regulators per gene", options: { fontSize: 10, color: C.textDark, breakLine: true } },
|
| 566 |
+
{ text: "99%+ values are noise!", options: { fontSize: 11, color: C.coral, bold: true, breakLine: true } },
|
| 567 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 568 |
+
{ text: "Evidence: latent loss ~ 1.12", options: { fontSize: 10, color: C.textMid, breakLine: true } },
|
| 569 |
+
{ text: " >> expr loss ~ 0.019", options: { fontSize: 10, color: C.textMid } },
|
| 570 |
+
], { x: 0.75, y: 0.95, w: 3.9, h: 1.9, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 571 |
+
|
| 572 |
+
// Solution 1 (top-right)
|
| 573 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.2, y: 0.9, w: 4.3, h: 2.0, fill: { color: C.white }, shadow: cardShadow() });
|
| 574 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.2, y: 0.9, w: 0.07, h: 2.0, fill: { color: C.accent3 } });
|
| 575 |
+
s.addText([
|
| 576 |
+
{ text: "Solution: Sparse Top-K", options: { bold: true, fontSize: 12, color: C.accent3, breakLine: true } },
|
| 577 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 578 |
+
{ text: "Per gene: keep only K=30 largest |delta|", options: { fontSize: 10, color: C.textDark, breakLine: true } },
|
| 579 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 580 |
+
{ text: "delta_attn (GxG) 25M values", options: { fontSize: 9.5, fontFace: "Consolas", color: C.coral, breakLine: true } },
|
| 581 |
+
{ text: " -> top-K sparsification", options: { fontSize: 9.5, fontFace: "Consolas", color: C.textMid, breakLine: true } },
|
| 582 |
+
{ text: "sparse_delta (Gx30) filter 99.4%", options: { fontSize: 9.5, fontFace: "Consolas", color: C.accent3, breakLine: true } },
|
| 583 |
+
{ text: " -> x gene_emb = (G,512)", options: { fontSize: 9.5, fontFace: "Consolas", color: C.accent3 } },
|
| 584 |
+
], { x: 5.45, y: 0.95, w: 3.9, h: 1.9, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 585 |
+
|
| 586 |
+
// Challenge 2 (bottom-left)
|
| 587 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 3.2, w: 4.3, h: 1.2, fill: { color: C.white }, shadow: cardShadow() });
|
| 588 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 3.2, w: 0.07, h: 1.2, fill: { color: C.coral } });
|
| 589 |
+
s.addText([
|
| 590 |
+
{ text: "Challenge 2: 512-d Latent Too Hard", options: { bold: true, fontSize: 12, color: C.coral, breakLine: true } },
|
| 591 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 592 |
+
{ text: "(G,512) = 2.5M-dim velocity field per step", options: { fontSize: 10, color: C.textDark, breakLine: true } },
|
| 593 |
+
{ text: "Ablation: dim 512->1: loss 1.1 -> 0.5-0.7", options: { fontSize: 10, fontFace: "Consolas", color: C.textDark } },
|
| 594 |
+
], { x: 0.75, y: 3.25, w: 3.9, h: 1.1, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 595 |
+
|
| 596 |
+
// Solution 2 (bottom-right)
|
| 597 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.2, y: 3.2, w: 4.3, h: 1.2, fill: { color: C.white }, shadow: cardShadow() });
|
| 598 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 5.2, y: 3.2, w: 0.07, h: 1.2, fill: { color: C.accent3 } });
|
| 599 |
+
s.addText([
|
| 600 |
+
{ text: "Solution: PCA Reduction", options: { bold: true, fontSize: 12, color: C.accent3, breakLine: true } },
|
| 601 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 602 |
+
{ text: "sparse_delta x pca_basis -> (G, 64)", options: { fontSize: 10, fontFace: "Consolas", color: C.textDark, breakLine: true } },
|
| 603 |
+
{ text: "Keep principal directions, 8x reduction", options: { fontSize: 10, color: C.textMid } },
|
| 604 |
+
], { x: 5.45, y: 3.25, w: 3.9, h: 1.1, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 605 |
+
}
|
| 606 |
+
|
| 607 |
+
// ============================================================
|
| 608 |
+
// SLIDE 15: Section — Summary
|
| 609 |
+
// ============================================================
|
| 610 |
+
slideNum++;
|
| 611 |
+
addDividerSlide("6. Summary & Future Work", "Validating the biological hypothesis", slideNum);
|
| 612 |
+
|
| 613 |
+
// ============================================================
|
| 614 |
+
// SLIDE 16: Summary + Future Experiment
|
| 615 |
+
// ============================================================
|
| 616 |
+
slideNum++;
|
| 617 |
+
{
|
| 618 |
+
const s = addContentSlide("Summary & Key Future Experiment", slideNum);
|
| 619 |
+
|
| 620 |
+
// Core contribution
|
| 621 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 0.85, w: 9.0, h: 0.9, fill: { color: C.navy } });
|
| 622 |
+
s.addText([
|
| 623 |
+
{ text: "Core contribution: ", options: { bold: true, color: C.gold, fontSize: 12 } },
|
| 624 |
+
{ text: "Not architectural improvement -- biological mechanism-driven modeling", options: { color: C.white, fontSize: 12, breakLine: true } },
|
| 625 |
+
{ text: "", options: { breakLine: true, fontSize: 3 } },
|
| 626 |
+
{ text: "Existing: Pert -> [black box] -> Expr ", options: { fontSize: 10, fontFace: "Consolas", color: C.midGray } },
|
| 627 |
+
{ text: "Ours: Pert -> GRN rewiring -> Expr", options: { fontSize: 10, fontFace: "Consolas", color: C.subtitleOnDark } },
|
| 628 |
+
], { x: 0.7, y: 0.88, w: 8.6, h: 0.85, fontFace: "Calibri", valign: "middle", margin: 0 });
|
| 629 |
+
|
| 630 |
+
// Future experiment
|
| 631 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 2.0, w: 9.0, h: 2.3, fill: { color: C.white }, shadow: cardShadow() });
|
| 632 |
+
s.addText([
|
| 633 |
+
{ text: "Key Future Experiment: Does inference order matter?", options: { bold: true, fontSize: 14, color: C.teal, breakLine: true } },
|
| 634 |
+
{ text: "", options: { breakLine: true, fontSize: 4 } },
|
| 635 |
+
{ text: "Train with random t1, t2 (no cascade). Compare inference orders:", options: { fontSize: 11, color: C.textDark } },
|
| 636 |
+
], { x: 0.7, y: 2.05, w: 8.6, h: 0.7, fontFace: "Calibri", valign: "top", margin: 0 });
|
| 637 |
+
|
| 638 |
+
const rows = [
|
| 639 |
+
{ order: "GRN first -> Expr", meaning: "Understand regulation, then predict", expected: "Best", bg: "D1FAE5", color: C.accent3 },
|
| 640 |
+
{ order: "Expr first -> GRN", meaning: "Predict first, understand after", expected: "Suboptimal", bg: "FEF3C7", color: C.accent2 },
|
| 641 |
+
{ order: "Simultaneous", meaning: "No explicit order", expected: "Worst", bg: "FEE2E2", color: C.coral },
|
| 642 |
+
];
|
| 643 |
+
rows.forEach((r, i) => {
|
| 644 |
+
const ry = 2.85 + i * 0.45;
|
| 645 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.8, y: ry, w: 8.4, h: 0.38, fill: { color: r.bg } });
|
| 646 |
+
s.addText(r.order, {
|
| 647 |
+
x: 0.9, y: ry, w: 2.8, h: 0.38,
|
| 648 |
+
fontSize: 11, fontFace: "Consolas", color: C.textDark, bold: true, valign: "middle", margin: 0,
|
| 649 |
+
});
|
| 650 |
+
s.addText(r.meaning, {
|
| 651 |
+
x: 3.8, y: ry, w: 3.2, h: 0.38,
|
| 652 |
+
fontSize: 10, fontFace: "Calibri", color: C.textMid, valign: "middle", margin: 0,
|
| 653 |
+
});
|
| 654 |
+
s.addText(r.expected, {
|
| 655 |
+
x: 7.2, y: ry, w: 1.8, h: 0.38,
|
| 656 |
+
fontSize: 12, fontFace: "Calibri", color: r.color, bold: true, valign: "middle", align: "center", margin: 0,
|
| 657 |
+
});
|
| 658 |
+
});
|
| 659 |
+
|
| 660 |
+
// Hypothesis
|
| 661 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.5, y: 4.55, w: 9.0, h: 0.5, fill: { color: C.navy } });
|
| 662 |
+
s.addText([
|
| 663 |
+
{ text: "Hypothesis: ", options: { bold: true, color: C.gold, fontSize: 12 } },
|
| 664 |
+
{ text: "Understanding GRN changes is a prerequisite for expression prediction, not a byproduct.", options: { color: C.white, fontSize: 12 } },
|
| 665 |
+
], { x: 0.7, y: 4.57, w: 8.6, h: 0.46, fontFace: "Calibri", valign: "middle", margin: 0 });
|
| 666 |
+
}
|
| 667 |
+
|
| 668 |
+
// ============================================================
|
| 669 |
+
// SLIDE 17: Closing
|
| 670 |
+
// ============================================================
|
| 671 |
+
slideNum++;
|
| 672 |
+
{
|
| 673 |
+
const s = pres.addSlide();
|
| 674 |
+
s.background = { color: C.navy };
|
| 675 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0, y: 0, w: 10, h: 0.08, fill: { color: C.seafoam } });
|
| 676 |
+
|
| 677 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.7, y: 1.5, w: 1.0, h: 0.06, fill: { color: C.gold } });
|
| 678 |
+
|
| 679 |
+
s.addText("Takeaway", {
|
| 680 |
+
x: 0.7, y: 1.7, w: 8.6, h: 0.5,
|
| 681 |
+
fontSize: 18, fontFace: "Georgia", color: C.white, bold: true, margin: 0,
|
| 682 |
+
});
|
| 683 |
+
|
| 684 |
+
s.addText("Use scGPT attention-delta to explicitly extract perturbation-induced GRN changes, and through cascaded flow matching, force the model to \"first understand how GRN changes, then predict how expression changes\" -- embedding the biological prior that perturbation propagates through GRN into the generative model's inference process.", {
|
| 685 |
+
x: 0.7, y: 2.4, w: 8.6, h: 2.0,
|
| 686 |
+
fontSize: 16, fontFace: "Georgia", color: C.white, lineSpacingMultiple: 1.5, margin: 0,
|
| 687 |
+
});
|
| 688 |
+
|
| 689 |
+
s.addShape(pres.shapes.RECTANGLE, { x: 0.7, y: 4.6, w: 2.5, h: 0.02, fill: { color: C.midGray } });
|
| 690 |
+
s.addText("Thank you!", {
|
| 691 |
+
x: 0.7, y: 4.75, w: 8.6, h: 0.45,
|
| 692 |
+
fontSize: 16, fontFace: "Georgia", color: C.white, bold: true, margin: 0,
|
| 693 |
+
});
|
| 694 |
+
addSlideNum(s, slideNum);
|
| 695 |
+
}
|
| 696 |
+
|
| 697 |
+
// === Save ===
|
| 698 |
+
const outPath = "/home/hp250092/ku50001222/qian/aivc/lfj/Report/GRN_CCFM_group_meeting.pptx";
|
| 699 |
+
pres.writeFile({ fileName: outPath }).then(() => {
|
| 700 |
+
console.log("Saved to: " + outPath);
|
| 701 |
+
}).catch(err => {
|
| 702 |
+
console.error("Error:", err);
|
| 703 |
+
});
|
Report/week10/GRN_Progress_Report.pdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c202c96142417b70f03e2f2df5df8756e82b87b20cdc5cc260442d0b3973d425
|
| 3 |
+
size 228576
|
Report/week10/GRN_Progress_Report.pptx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:784e7892790d73089573d185c67036ff63ed5cf9ee786620e169ad5d39ab8bba
|
| 3 |
+
size 668124
|
Report/week10/gen_pptx.js
ADDED
|
@@ -0,0 +1,874 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
const pptxgen = require("pptxgenjs");
|
| 2 |
+
|
| 3 |
+
const pres = new pptxgen();
|
| 4 |
+
pres.layout = "LAYOUT_16x9";
|
| 5 |
+
pres.author = "Qian";
|
| 6 |
+
pres.title = "GRN-Guided Perturbation Prediction";
|
| 7 |
+
|
| 8 |
+
// ── Design Tokens ──────────────────────────────────────────────────
|
| 9 |
+
const C = {
|
| 10 |
+
primary: "990011", // cherry red
|
| 11 |
+
primaryLt: "B8001A", // lighter red
|
| 12 |
+
accent: "2F3C7E", // navy accent
|
| 13 |
+
dark: "1A1A2E", // near-black for title slides
|
| 14 |
+
white: "FFFFFF",
|
| 15 |
+
offWhite: "FCF6F5",
|
| 16 |
+
lightGray: "F5F0EE",
|
| 17 |
+
midGray: "AAAAAA",
|
| 18 |
+
darkText: "2D2D2D",
|
| 19 |
+
bodyText: "3A3A3A",
|
| 20 |
+
tableHead: "990011",
|
| 21 |
+
tableAlt: "FDF2F2",
|
| 22 |
+
green: "2E7D32",
|
| 23 |
+
red: "C62828",
|
| 24 |
+
};
|
| 25 |
+
const FONT_H = "Georgia";
|
| 26 |
+
const FONT_B = "Calibri";
|
| 27 |
+
|
| 28 |
+
// ── Helper: section divider bar (top) ──────────────────────────────
|
| 29 |
+
function addTopBar(slide) {
|
| 30 |
+
slide.addShape(pres.shapes.RECTANGLE, {
|
| 31 |
+
x: 0, y: 0, w: 10, h: 0.06,
|
| 32 |
+
fill: { color: C.primary },
|
| 33 |
+
});
|
| 34 |
+
}
|
| 35 |
+
|
| 36 |
+
// ── Helper: slide number ───────────────────────────────────────────
|
| 37 |
+
function addSlideNum(slide, num) {
|
| 38 |
+
slide.addText(String(num), {
|
| 39 |
+
x: 9.2, y: 5.15, w: 0.6, h: 0.35,
|
| 40 |
+
fontSize: 10, fontFace: FONT_B, color: C.midGray, align: "right",
|
| 41 |
+
});
|
| 42 |
+
}
|
| 43 |
+
|
| 44 |
+
// ── Helper: content slide title ────────────────────────────────────
|
| 45 |
+
function addTitle(slide, title, num) {
|
| 46 |
+
addTopBar(slide);
|
| 47 |
+
slide.addText(title, {
|
| 48 |
+
x: 0.6, y: 0.25, w: 8.8, h: 0.55,
|
| 49 |
+
fontSize: 28, fontFace: FONT_H, color: C.primary, bold: true, margin: 0,
|
| 50 |
+
});
|
| 51 |
+
// thin separator
|
| 52 |
+
slide.addShape(pres.shapes.LINE, {
|
| 53 |
+
x: 0.6, y: 0.85, w: 8.8, h: 0,
|
| 54 |
+
line: { color: C.primary, width: 1.2 },
|
| 55 |
+
});
|
| 56 |
+
addSlideNum(slide, num);
|
| 57 |
+
}
|
| 58 |
+
|
| 59 |
+
// ── Helper: bullet list ────────────────────────────────────────────
|
| 60 |
+
function bullets(items, opts = {}) {
|
| 61 |
+
return items.map((t, i) => ({
|
| 62 |
+
text: typeof t === "string" ? t : t.text,
|
| 63 |
+
options: {
|
| 64 |
+
bullet: { code: "2022" },
|
| 65 |
+
breakLine: i < items.length - 1,
|
| 66 |
+
fontSize: opts.fontSize || 14,
|
| 67 |
+
fontFace: FONT_B,
|
| 68 |
+
color: opts.color || C.bodyText,
|
| 69 |
+
bold: (typeof t === "object" && t.bold) || false,
|
| 70 |
+
indentLevel: (typeof t === "object" && t.indent) || 0,
|
| 71 |
+
paraSpaceAfter: 6,
|
| 72 |
+
},
|
| 73 |
+
}));
|
| 74 |
+
}
|
| 75 |
+
|
| 76 |
+
// ── Helper: table with standard styling ────────────────────────────
|
| 77 |
+
function styledTable(headerRow, dataRows, opts = {}) {
|
| 78 |
+
const hdrCells = headerRow.map((h) => ({
|
| 79 |
+
text: h, options: {
|
| 80 |
+
bold: true, color: C.white, fill: { color: C.tableHead },
|
| 81 |
+
fontSize: 11, fontFace: FONT_B, align: "center", valign: "middle",
|
| 82 |
+
},
|
| 83 |
+
}));
|
| 84 |
+
const bodyRows = dataRows.map((row, ri) =>
|
| 85 |
+
row.map((cell) => {
|
| 86 |
+
const isObj = typeof cell === "object" && cell !== null;
|
| 87 |
+
return {
|
| 88 |
+
text: isObj ? cell.text : String(cell),
|
| 89 |
+
options: {
|
| 90 |
+
fontSize: 11, fontFace: FONT_B, color: C.darkText,
|
| 91 |
+
fill: { color: ri % 2 === 0 ? C.white : C.tableAlt },
|
| 92 |
+
align: isObj && cell.align ? cell.align : "center",
|
| 93 |
+
valign: "middle",
|
| 94 |
+
bold: isObj && cell.bold ? true : false,
|
| 95 |
+
},
|
| 96 |
+
};
|
| 97 |
+
})
|
| 98 |
+
);
|
| 99 |
+
return { rows: [hdrCells, ...bodyRows], opts };
|
| 100 |
+
}
|
| 101 |
+
|
| 102 |
+
// ====================================================================
|
| 103 |
+
// SLIDE 1 — Title
|
| 104 |
+
// ====================================================================
|
| 105 |
+
{
|
| 106 |
+
const s = pres.addSlide();
|
| 107 |
+
s.background = { color: C.dark };
|
| 108 |
+
// accent bar left
|
| 109 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 110 |
+
x: 0, y: 0, w: 0.12, h: 5.625, fill: { color: C.primary },
|
| 111 |
+
});
|
| 112 |
+
s.addText("GRN-Guided Perturbation Prediction", {
|
| 113 |
+
x: 0.7, y: 1.2, w: 8.6, h: 1.2,
|
| 114 |
+
fontSize: 36, fontFace: FONT_H, color: C.white, bold: true, margin: 0,
|
| 115 |
+
});
|
| 116 |
+
s.addText("From Cascaded Flow Matching to RegFM", {
|
| 117 |
+
x: 0.7, y: 2.4, w: 8.6, h: 0.6,
|
| 118 |
+
fontSize: 20, fontFace: FONT_B, color: C.primaryLt, margin: 0,
|
| 119 |
+
});
|
| 120 |
+
s.addShape(pres.shapes.LINE, {
|
| 121 |
+
x: 0.7, y: 3.2, w: 3, h: 0,
|
| 122 |
+
line: { color: C.primary, width: 2 },
|
| 123 |
+
});
|
| 124 |
+
s.addText("Weekly Research Progress Report | 2026-03-29", {
|
| 125 |
+
x: 0.7, y: 3.5, w: 8.6, h: 0.4,
|
| 126 |
+
fontSize: 14, fontFace: FONT_B, color: "CCCCCC", margin: 0,
|
| 127 |
+
});
|
| 128 |
+
}
|
| 129 |
+
|
| 130 |
+
// ====================================================================
|
| 131 |
+
// SLIDE 2 — Task Overview
|
| 132 |
+
// ====================================================================
|
| 133 |
+
{
|
| 134 |
+
const s = pres.addSlide();
|
| 135 |
+
s.background = { color: C.white };
|
| 136 |
+
addTitle(s, "Task: Single-Cell Perturbation Prediction", 2);
|
| 137 |
+
|
| 138 |
+
// Left column — description
|
| 139 |
+
s.addText(bullets([
|
| 140 |
+
"Input: control expression + perturbed gene(s)",
|
| 141 |
+
"Output: post-perturbation expression",
|
| 142 |
+
"No cell-level pairing (destructive assay)",
|
| 143 |
+
"Eval: DE overlap, Pearson, MSE, etc.",
|
| 144 |
+
]), { x: 0.6, y: 1.15, w: 4.5, h: 2.2 });
|
| 145 |
+
|
| 146 |
+
// Right column — dataset card
|
| 147 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 148 |
+
x: 5.5, y: 1.15, w: 4.0, h: 2.3,
|
| 149 |
+
fill: { color: C.lightGray },
|
| 150 |
+
});
|
| 151 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 152 |
+
x: 5.5, y: 1.15, w: 4.0, h: 0.4,
|
| 153 |
+
fill: { color: C.primary },
|
| 154 |
+
});
|
| 155 |
+
s.addText("Norman Dataset", {
|
| 156 |
+
x: 5.5, y: 1.15, w: 4.0, h: 0.4,
|
| 157 |
+
fontSize: 13, fontFace: FONT_B, color: C.white, bold: true, align: "center", valign: "middle",
|
| 158 |
+
});
|
| 159 |
+
s.addText(bullets([
|
| 160 |
+
"~9,000 cells x 5,000 HVG",
|
| 161 |
+
"105 CRISPR perturbations (KO + OE)",
|
| 162 |
+
"39 held-out test perturbations",
|
| 163 |
+
"Fold-1 split (additive)",
|
| 164 |
+
], { fontSize: 12 }), { x: 5.7, y: 1.65, w: 3.6, h: 1.7 });
|
| 165 |
+
|
| 166 |
+
// Formalization box
|
| 167 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 168 |
+
x: 0.6, y: 3.7, w: 8.8, h: 1.4,
|
| 169 |
+
fill: { color: C.offWhite },
|
| 170 |
+
});
|
| 171 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 172 |
+
x: 0.6, y: 3.7, w: 0.08, h: 1.4,
|
| 173 |
+
fill: { color: C.primary },
|
| 174 |
+
});
|
| 175 |
+
s.addText([
|
| 176 |
+
{ text: "Formal Definition", options: { bold: true, fontSize: 14, fontFace: FONT_H, color: C.primary, breakLine: true, paraSpaceAfter: 4 } },
|
| 177 |
+
{ text: "Given: x_ctrl in R^G (control expression), p in {gene_1, gene_2} (perturbation)", options: { fontSize: 12, fontFace: FONT_B, color: C.bodyText, breakLine: true, paraSpaceAfter: 2 } },
|
| 178 |
+
{ text: "Predict: x_pert in R^G (post-perturbation expression)", options: { fontSize: 12, fontFace: FONT_B, color: C.bodyText } },
|
| 179 |
+
], { x: 0.9, y: 3.8, w: 8.3, h: 1.2 });
|
| 180 |
+
}
|
| 181 |
+
|
| 182 |
+
// ====================================================================
|
| 183 |
+
// SLIDE 3 — Baseline: scDFM
|
| 184 |
+
// ====================================================================
|
| 185 |
+
{
|
| 186 |
+
const s = pres.addSlide();
|
| 187 |
+
s.background = { color: C.white };
|
| 188 |
+
addTitle(s, "Baseline: scDFM (Flow Matching)", 3);
|
| 189 |
+
|
| 190 |
+
s.addText(bullets([
|
| 191 |
+
"Learns velocity field v(x, t) via flow matching",
|
| 192 |
+
"Transports control distribution to perturbed",
|
| 193 |
+
"DiffPerceiverBlock backbone (d=128, 4 layers)",
|
| 194 |
+
"ODE solver: RK4, 20 steps",
|
| 195 |
+
]), { x: 0.6, y: 1.15, w: 5.0, h: 2.0 });
|
| 196 |
+
|
| 197 |
+
// Metrics callout cards
|
| 198 |
+
const metrics = [
|
| 199 |
+
{ label: "Pearson Delta", value: "0.866" },
|
| 200 |
+
{ label: "MSE", value: "0.0032" },
|
| 201 |
+
{ label: "DE Direction", value: "93.7%" },
|
| 202 |
+
{ label: "Discrimination", value: "0.980" },
|
| 203 |
+
];
|
| 204 |
+
metrics.forEach((m, i) => {
|
| 205 |
+
const mx = 0.6 + i * 2.3;
|
| 206 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 207 |
+
x: mx, y: 2.9, w: 2.1, h: 1.5,
|
| 208 |
+
fill: { color: C.offWhite },
|
| 209 |
+
});
|
| 210 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 211 |
+
x: mx, y: 2.9, w: 2.1, h: 0.06,
|
| 212 |
+
fill: { color: C.primary },
|
| 213 |
+
});
|
| 214 |
+
s.addText(m.value, {
|
| 215 |
+
x: mx, y: 3.1, w: 2.1, h: 0.6,
|
| 216 |
+
fontSize: 26, fontFace: FONT_H, color: C.primary, bold: true, align: "center", margin: 0,
|
| 217 |
+
});
|
| 218 |
+
s.addText(m.label, {
|
| 219 |
+
x: mx, y: 3.7, w: 2.1, h: 0.4,
|
| 220 |
+
fontSize: 12, fontFace: FONT_B, color: "777777", align: "center", margin: 0,
|
| 221 |
+
});
|
| 222 |
+
});
|
| 223 |
+
|
| 224 |
+
// Limitation note
|
| 225 |
+
s.addText([
|
| 226 |
+
{ text: "Limitation: ", options: { bold: true, fontSize: 13, color: C.red } },
|
| 227 |
+
{ text: "Genes treated as unstructured vector; no GRN modeling", options: { fontSize: 13, color: C.bodyText } },
|
| 228 |
+
], { x: 0.6, y: 4.7, w: 8.8, h: 0.3, fontFace: FONT_B });
|
| 229 |
+
}
|
| 230 |
+
|
| 231 |
+
// ====================================================================
|
| 232 |
+
// SLIDE 4 — GRN-CCFM: Cascaded Approach
|
| 233 |
+
// ====================================================================
|
| 234 |
+
{
|
| 235 |
+
const s = pres.addSlide();
|
| 236 |
+
s.background = { color: C.white };
|
| 237 |
+
addTitle(s, "GRN-CCFM: Cascaded Approach", 4);
|
| 238 |
+
|
| 239 |
+
// Core idea
|
| 240 |
+
s.addText([
|
| 241 |
+
{ text: "Core Idea", options: { bold: true, fontSize: 16, fontFace: FONT_H, color: C.primary, breakLine: true, paraSpaceAfter: 6 } },
|
| 242 |
+
{ text: "Extract Attention-Delta from scGPT to capture GRN change, then use LatentForcing cascade to jointly generate GRN features and gene expression.", options: { fontSize: 13, fontFace: FONT_B, color: C.bodyText } },
|
| 243 |
+
], { x: 0.6, y: 1.15, w: 8.8, h: 1.0 });
|
| 244 |
+
|
| 245 |
+
// Three pillars
|
| 246 |
+
const pillars = [
|
| 247 |
+
{ title: "Attention-Delta", items: ["delta_attn = attn_tgt - attn_ctrl", "Captures regulatory change", "Sparse G x G matrix (~0.6%)"] },
|
| 248 |
+
{ title: "Cascaded Training", items: ["40% steps: latent flow only", "60% steps: expression flow only", "Two-stage ODE at inference"] },
|
| 249 |
+
{ title: "Architecture Fix", items: ["d_model: 128 -> 512", "Missing gene mask (7 sites)", "scGPT vocab alignment"] },
|
| 250 |
+
];
|
| 251 |
+
pillars.forEach((p, i) => {
|
| 252 |
+
const px = 0.6 + i * 3.1;
|
| 253 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 254 |
+
x: px, y: 2.4, w: 2.8, h: 2.8,
|
| 255 |
+
fill: { color: C.offWhite },
|
| 256 |
+
});
|
| 257 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 258 |
+
x: px, y: 2.4, w: 2.8, h: 0.45,
|
| 259 |
+
fill: { color: C.primary },
|
| 260 |
+
});
|
| 261 |
+
s.addText(p.title, {
|
| 262 |
+
x: px, y: 2.4, w: 2.8, h: 0.45,
|
| 263 |
+
fontSize: 13, fontFace: FONT_B, color: C.white, bold: true, align: "center", valign: "middle",
|
| 264 |
+
});
|
| 265 |
+
s.addText(bullets(p.items, { fontSize: 11 }), {
|
| 266 |
+
x: px + 0.15, y: 3.0, w: 2.5, h: 2.0,
|
| 267 |
+
});
|
| 268 |
+
});
|
| 269 |
+
}
|
| 270 |
+
|
| 271 |
+
// ====================================================================
|
| 272 |
+
// SLIDE 5 — Cascaded Variants Overview
|
| 273 |
+
// ====================================================================
|
| 274 |
+
{
|
| 275 |
+
const s = pres.addSlide();
|
| 276 |
+
s.background = { color: C.white };
|
| 277 |
+
addTitle(s, "Cascaded Variants: Design Space", 5);
|
| 278 |
+
|
| 279 |
+
const tbl = styledTable(
|
| 280 |
+
["Variant", "Latent Dim", "Agg. Method", "delta_topk", "Key Idea"],
|
| 281 |
+
[
|
| 282 |
+
["grn_att_only", "128 (bilinear)", "Bilinear head", "30", "Attention only, no SVD"],
|
| 283 |
+
["grn_svd", "128", "SVD dictionary", "30", "Fixed SVD basis"],
|
| 284 |
+
["grn_svd_cross", "128", "SVD + cross-attn", "30", "Learnable SVD queries"],
|
| 285 |
+
["grn_dense4", "4", "Multi-stats", "30", "Low-dim dense features"],
|
| 286 |
+
["grn_scalar", "1", "signed_L2 + norm", "100", "Scalar latent per gene"],
|
| 287 |
+
["dim1_ablation", "1", "Slice scGPT[0]", "30", "Ablation: 512d -> 1d"],
|
| 288 |
+
]
|
| 289 |
+
);
|
| 290 |
+
s.addTable(tbl.rows, {
|
| 291 |
+
x: 0.5, y: 1.15, w: 9.0,
|
| 292 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 293 |
+
colW: [1.5, 1.3, 1.5, 1.0, 3.7],
|
| 294 |
+
rowH: [0.38, 0.35, 0.35, 0.35, 0.35, 0.35, 0.35],
|
| 295 |
+
autoPage: false,
|
| 296 |
+
});
|
| 297 |
+
|
| 298 |
+
// Shared config note
|
| 299 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 300 |
+
x: 0.5, y: 3.85, w: 9.0, h: 1.3,
|
| 301 |
+
fill: { color: C.offWhite },
|
| 302 |
+
});
|
| 303 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 304 |
+
x: 0.5, y: 3.85, w: 0.08, h: 1.3,
|
| 305 |
+
fill: { color: C.accent },
|
| 306 |
+
});
|
| 307 |
+
s.addText([
|
| 308 |
+
{ text: "Shared Config", options: { bold: true, fontSize: 13, fontFace: FONT_H, color: C.accent, breakLine: true, paraSpaceAfter: 4 } },
|
| 309 |
+
], { x: 0.8, y: 3.95, w: 8.5, h: 0.3 });
|
| 310 |
+
s.addText(bullets([
|
| 311 |
+
"d_model=128, nlayers=4, nhead=8, lr=5e-5, EMA=0.9999",
|
| 312 |
+
"Cascaded: choose_latent_p=0.4, latent_weight=1.0",
|
| 313 |
+
"Inference: RK4 ODE, 20 steps each stage",
|
| 314 |
+
], { fontSize: 11 }), { x: 0.8, y: 4.25, w: 8.5, h: 0.8 });
|
| 315 |
+
}
|
| 316 |
+
|
| 317 |
+
// ====================================================================
|
| 318 |
+
// SLIDE 6 — Cascaded Results
|
| 319 |
+
// ====================================================================
|
| 320 |
+
{
|
| 321 |
+
const s = pres.addSlide();
|
| 322 |
+
s.background = { color: C.white };
|
| 323 |
+
addTitle(s, "Cascaded Results: Evaluation Metrics", 6);
|
| 324 |
+
|
| 325 |
+
const tbl = styledTable(
|
| 326 |
+
["Model", "Pearson Delta", "MSE", "DE Direction", "Discrim."],
|
| 327 |
+
[
|
| 328 |
+
[{ text: "scDFM Baseline", bold: true }, { text: "0.866", bold: true }, { text: "0.0032", bold: true }, { text: "0.937", bold: true }, { text: "0.980", bold: true }],
|
| 329 |
+
[{ text: "dim1_ablation", bold: true }, { text: "0.752", bold: true }, "0.0059", "0.878", "0.914"],
|
| 330 |
+
["grn_dense4", "0.122", "0.020", "0.780", "0.521"],
|
| 331 |
+
["grn_scalar (dtk100)", "0.087", "0.021", "0.793", "0.534"],
|
| 332 |
+
["grn_scalar (bs48)", "0.068", "0.026", "0.771", "0.533"],
|
| 333 |
+
["grn_att_only", "-0.097", "0.602", "0.747", "0.552"],
|
| 334 |
+
["grn_svd / svd_cross", "-0.096", "0.575", "0.746", "0.492"],
|
| 335 |
+
]
|
| 336 |
+
);
|
| 337 |
+
s.addTable(tbl.rows, {
|
| 338 |
+
x: 0.5, y: 1.15, w: 9.0,
|
| 339 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 340 |
+
colW: [2.2, 1.7, 1.3, 1.6, 1.4],
|
| 341 |
+
rowH: [0.4, 0.36, 0.36, 0.36, 0.36, 0.36, 0.36, 0.36],
|
| 342 |
+
autoPage: false,
|
| 343 |
+
});
|
| 344 |
+
|
| 345 |
+
// Insight box
|
| 346 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 347 |
+
x: 0.5, y: 4.2, w: 9.0, h: 1.05,
|
| 348 |
+
fill: { color: "FFF5F5" },
|
| 349 |
+
});
|
| 350 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 351 |
+
x: 0.5, y: 4.2, w: 0.08, h: 1.05,
|
| 352 |
+
fill: { color: C.red },
|
| 353 |
+
});
|
| 354 |
+
s.addText(bullets([
|
| 355 |
+
{ text: "Only dim1 (d=1) approaches baseline; all high-dim cascaded variants fail", bold: true },
|
| 356 |
+
"High-dim latent generation is the fundamental bottleneck",
|
| 357 |
+
"grn_att_only / grn_svd: negative Pearson, MSE > 0.5",
|
| 358 |
+
], { fontSize: 12 }), { x: 0.8, y: 4.25, w: 8.5, h: 0.95 });
|
| 359 |
+
}
|
| 360 |
+
|
| 361 |
+
// ====================================================================
|
| 362 |
+
// SLIDE 7 — Failure Analysis
|
| 363 |
+
// ====================================================================
|
| 364 |
+
{
|
| 365 |
+
const s = pres.addSlide();
|
| 366 |
+
s.background = { color: C.white };
|
| 367 |
+
addTitle(s, "Failure Analysis: Why Cascaded Fails", 7);
|
| 368 |
+
|
| 369 |
+
// Left: problem
|
| 370 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 371 |
+
x: 0.5, y: 1.15, w: 4.3, h: 3.5,
|
| 372 |
+
fill: { color: "FFF5F5" },
|
| 373 |
+
});
|
| 374 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 375 |
+
x: 0.5, y: 1.15, w: 4.3, h: 0.45,
|
| 376 |
+
fill: { color: C.red },
|
| 377 |
+
});
|
| 378 |
+
s.addText("Root Cause", {
|
| 379 |
+
x: 0.5, y: 1.15, w: 4.3, h: 0.45,
|
| 380 |
+
fontSize: 14, fontFace: FONT_B, color: C.white, bold: true, align: "center", valign: "middle",
|
| 381 |
+
});
|
| 382 |
+
s.addText(bullets([
|
| 383 |
+
"Latent loss stuck at ~1.0-2.0",
|
| 384 |
+
"Target: sparse G x G matrix (~0.6% non-zero)",
|
| 385 |
+
"Generating GRN is itself a hard problem",
|
| 386 |
+
"Decoupled training prevents joint optimization",
|
| 387 |
+
"Expression flow never benefits from GRN",
|
| 388 |
+
], { fontSize: 12 }), { x: 0.7, y: 1.8, w: 3.9, h: 2.5 });
|
| 389 |
+
|
| 390 |
+
// Right: evidence
|
| 391 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 392 |
+
x: 5.2, y: 1.15, w: 4.3, h: 3.5,
|
| 393 |
+
fill: { color: C.offWhite },
|
| 394 |
+
});
|
| 395 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 396 |
+
x: 5.2, y: 1.15, w: 4.3, h: 0.45,
|
| 397 |
+
fill: { color: C.green },
|
| 398 |
+
});
|
| 399 |
+
s.addText("Evidence from dim1 Ablation", {
|
| 400 |
+
x: 5.2, y: 1.15, w: 4.3, h: 0.45,
|
| 401 |
+
fontSize: 14, fontFace: FONT_B, color: C.white, bold: true, align: "center", valign: "middle",
|
| 402 |
+
});
|
| 403 |
+
s.addText(bullets([
|
| 404 |
+
"scgpt_dim: 512 -> 1",
|
| 405 |
+
"Latent loss converges normally",
|
| 406 |
+
"Pearson delta: 0.752 (vs baseline 0.866)",
|
| 407 |
+
"Confirms: high-dim target is the bottleneck",
|
| 408 |
+
"But dim=1 loses most GRN information",
|
| 409 |
+
], { fontSize: 12 }), { x: 5.4, y: 1.8, w: 3.9, h: 2.5 });
|
| 410 |
+
|
| 411 |
+
// Bottom conclusion
|
| 412 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 413 |
+
x: 0.5, y: 4.85, w: 9.0, h: 0.5,
|
| 414 |
+
fill: { color: C.dark },
|
| 415 |
+
});
|
| 416 |
+
s.addText("Conclusion: Cascaded generation of GRN features is a dead end. Need a new paradigm.", {
|
| 417 |
+
x: 0.5, y: 4.85, w: 9.0, h: 0.5,
|
| 418 |
+
fontSize: 13, fontFace: FONT_B, color: C.white, bold: true, align: "center", valign: "middle",
|
| 419 |
+
});
|
| 420 |
+
}
|
| 421 |
+
|
| 422 |
+
// ====================================================================
|
| 423 |
+
// SLIDE 8 — Paradigm Shift
|
| 424 |
+
// ====================================================================
|
| 425 |
+
{
|
| 426 |
+
const s = pres.addSlide();
|
| 427 |
+
s.background = { color: C.dark };
|
| 428 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 429 |
+
x: 0, y: 0, w: 0.12, h: 5.625, fill: { color: C.primary },
|
| 430 |
+
});
|
| 431 |
+
|
| 432 |
+
s.addText("Paradigm Shift", {
|
| 433 |
+
x: 0.7, y: 0.8, w: 8.6, h: 0.6,
|
| 434 |
+
fontSize: 32, fontFace: FONT_H, color: C.white, bold: true, margin: 0,
|
| 435 |
+
});
|
| 436 |
+
s.addText("From GRN Generation to Structural Supervision", {
|
| 437 |
+
x: 0.7, y: 1.4, w: 8.6, h: 0.5,
|
| 438 |
+
fontSize: 18, fontFace: FONT_B, color: C.primaryLt, margin: 0,
|
| 439 |
+
});
|
| 440 |
+
|
| 441 |
+
// Key insight box
|
| 442 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 443 |
+
x: 0.7, y: 2.3, w: 8.6, h: 1.2,
|
| 444 |
+
fill: { color: C.primary },
|
| 445 |
+
});
|
| 446 |
+
s.addText([
|
| 447 |
+
{ text: "Key Insight: ", options: { bold: true, fontSize: 15, color: C.white } },
|
| 448 |
+
{ text: "Delta-attention is ", options: { fontSize: 15, color: C.white } },
|
| 449 |
+
{ text: "privileged information", options: { bold: true, italic: true, fontSize: 15, color: C.white } },
|
| 450 |
+
{ text: " -- available at training (from source + target), absent at inference (only source).", options: { fontSize: 15, color: C.white } },
|
| 451 |
+
], { x: 1.0, y: 2.5, w: 8.0, h: 0.8, fontFace: FONT_B });
|
| 452 |
+
|
| 453 |
+
// Old vs New
|
| 454 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 455 |
+
x: 0.7, y: 3.8, w: 4.0, h: 1.2,
|
| 456 |
+
fill: { color: "2A2A42" },
|
| 457 |
+
});
|
| 458 |
+
s.addText([
|
| 459 |
+
{ text: "OLD: Cascaded", options: { bold: true, fontSize: 14, color: C.red, breakLine: true, paraSpaceAfter: 4 } },
|
| 460 |
+
{ text: "Generate GRN features (latent flow)", options: { fontSize: 13, color: "E8E8E8", breakLine: true, paraSpaceAfter: 2 } },
|
| 461 |
+
{ text: "-> Use for expression (expr flow)", options: { fontSize: 13, color: "E8E8E8" } },
|
| 462 |
+
], { x: 0.9, y: 3.9, w: 3.6, h: 1.0, fontFace: FONT_B });
|
| 463 |
+
|
| 464 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 465 |
+
x: 5.3, y: 3.8, w: 4.0, h: 1.2,
|
| 466 |
+
fill: { color: "2A2A42" },
|
| 467 |
+
});
|
| 468 |
+
s.addText([
|
| 469 |
+
{ text: "NEW: RegFM", options: { bold: true, fontSize: 14, color: "66BB6A", breakLine: true, paraSpaceAfter: 4 } },
|
| 470 |
+
{ text: "Embed GRN as structural bias", options: { fontSize: 13, color: "E8E8E8", breakLine: true, paraSpaceAfter: 2 } },
|
| 471 |
+
{ text: "-> Soft-supervise with delta_attn at train", options: { fontSize: 13, color: "E8E8E8" } },
|
| 472 |
+
], { x: 5.5, y: 3.9, w: 3.6, h: 1.0, fontFace: FONT_B });
|
| 473 |
+
|
| 474 |
+
addSlideNum(s, 8);
|
| 475 |
+
}
|
| 476 |
+
|
| 477 |
+
// ====================================================================
|
| 478 |
+
// SLIDE 9 — RegFM Architecture
|
| 479 |
+
// ====================================================================
|
| 480 |
+
{
|
| 481 |
+
const s = pres.addSlide();
|
| 482 |
+
s.background = { color: C.white };
|
| 483 |
+
addTitle(s, "RegFM: Architecture", 9);
|
| 484 |
+
|
| 485 |
+
// Equation box
|
| 486 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 487 |
+
x: 0.6, y: 1.15, w: 8.8, h: 0.7,
|
| 488 |
+
fill: { color: C.offWhite },
|
| 489 |
+
});
|
| 490 |
+
s.addText([
|
| 491 |
+
{ text: "v(x, t) = \u03B1 \u00B7 v", options: { fontFace: "Calibri", fontSize: 22, color: C.primary, bold: true } },
|
| 492 |
+
{ text: "reg", options: { fontFace: "Calibri", fontSize: 15, color: C.primary, bold: true } },
|
| 493 |
+
{ text: " + (1 \u2013 \u03B1) \u00B7 v", options: { fontFace: "Calibri", fontSize: 22, color: C.primary, bold: true } },
|
| 494 |
+
{ text: "int", options: { fontFace: "Calibri", fontSize: 15, color: C.primary, bold: true } },
|
| 495 |
+
], {
|
| 496 |
+
x: 0.6, y: 1.15, w: 8.8, h: 0.7,
|
| 497 |
+
align: "center", valign: "middle",
|
| 498 |
+
});
|
| 499 |
+
|
| 500 |
+
// Three component cards
|
| 501 |
+
const comps = [
|
| 502 |
+
{
|
| 503 |
+
title: "v_reg (Regulatory)",
|
| 504 |
+
color: C.primary,
|
| 505 |
+
items: [
|
| 506 |
+
"RegulatoryHead module",
|
| 507 |
+
"Q, K, V from backbone h",
|
| 508 |
+
"R = tanh(QK^T / sqrt(d_r))",
|
| 509 |
+
"v_reg = Linear(R @ V)",
|
| 510 |
+
"d_r = 32, params = 12K",
|
| 511 |
+
],
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
title: "v_int (Intrinsic)",
|
| 515 |
+
color: C.accent,
|
| 516 |
+
items: [
|
| 517 |
+
"Original ExprDecoder",
|
| 518 |
+
"3-layer MLP",
|
| 519 |
+
"Per-gene autonomous dynamics",
|
| 520 |
+
"No inter-gene interaction",
|
| 521 |
+
"Reused from scDFM baseline",
|
| 522 |
+
],
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
title: "Gate (alpha)",
|
| 526 |
+
color: C.green,
|
| 527 |
+
items: [
|
| 528 |
+
"VelocityGate module",
|
| 529 |
+
"Input: h + pert_emb + t_emb",
|
| 530 |
+
"MLP(384 -> 128 -> 1)",
|
| 531 |
+
"Init: bias=-3, alpha~0.05",
|
| 532 |
+
"Safe fallback to v_int",
|
| 533 |
+
],
|
| 534 |
+
},
|
| 535 |
+
];
|
| 536 |
+
comps.forEach((c, i) => {
|
| 537 |
+
const cx = 0.5 + i * 3.15;
|
| 538 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 539 |
+
x: cx, y: 2.15, w: 2.95, h: 3.0,
|
| 540 |
+
fill: { color: C.offWhite },
|
| 541 |
+
});
|
| 542 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 543 |
+
x: cx, y: 2.15, w: 2.95, h: 0.42,
|
| 544 |
+
fill: { color: c.color },
|
| 545 |
+
});
|
| 546 |
+
s.addText(c.title, {
|
| 547 |
+
x: cx, y: 2.15, w: 2.95, h: 0.42,
|
| 548 |
+
fontSize: 12, fontFace: FONT_B, color: C.white, bold: true, align: "center", valign: "middle",
|
| 549 |
+
});
|
| 550 |
+
s.addText(bullets(c.items, { fontSize: 10.5 }), {
|
| 551 |
+
x: cx + 0.1, y: 2.7, w: 2.75, h: 2.3,
|
| 552 |
+
});
|
| 553 |
+
});
|
| 554 |
+
}
|
| 555 |
+
|
| 556 |
+
// ====================================================================
|
| 557 |
+
// SLIDE 10 — RegFM Training & Loss
|
| 558 |
+
// ====================================================================
|
| 559 |
+
{
|
| 560 |
+
const s = pres.addSlide();
|
| 561 |
+
s.background = { color: C.white };
|
| 562 |
+
addTitle(s, "RegFM: Training & Loss Design", 10);
|
| 563 |
+
|
| 564 |
+
// Loss equation
|
| 565 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 566 |
+
x: 0.6, y: 1.15, w: 8.8, h: 0.6,
|
| 567 |
+
fill: { color: C.offWhite },
|
| 568 |
+
});
|
| 569 |
+
s.addText([
|
| 570 |
+
{ text: "L = L", options: { fontFace: "Calibri", fontSize: 20, color: C.primary, bold: true } },
|
| 571 |
+
{ text: "vel", options: { fontFace: "Calibri", fontSize: 14, color: C.primary, bold: true } },
|
| 572 |
+
{ text: " + \u03BB \u00B7 L", options: { fontFace: "Calibri", fontSize: 20, color: C.primary, bold: true } },
|
| 573 |
+
{ text: "reg", options: { fontFace: "Calibri", fontSize: 14, color: C.primary, bold: true } },
|
| 574 |
+
{ text: " + \u03B3 \u00B7 L", options: { fontFace: "Calibri", fontSize: 20, color: C.primary, bold: true } },
|
| 575 |
+
{ text: "mmd", options: { fontFace: "Calibri", fontSize: 14, color: C.primary, bold: true } },
|
| 576 |
+
], {
|
| 577 |
+
x: 0.6, y: 1.15, w: 8.8, h: 0.6,
|
| 578 |
+
align: "center", valign: "middle",
|
| 579 |
+
});
|
| 580 |
+
|
| 581 |
+
// Loss details table
|
| 582 |
+
const tbl = styledTable(
|
| 583 |
+
["Loss Term", "Target", "Weight", "Description"],
|
| 584 |
+
[
|
| 585 |
+
["L_vel", "v_target = x1 - eps", "1.0", "Standard flow matching MSE"],
|
| 586 |
+
["L_reg", "delta_attention", "0.1", "R_theta aligned with GRN ground truth"],
|
| 587 |
+
["L_mmd", "Distribution matching", "0.5", "Sliced Wasserstein / MMD"],
|
| 588 |
+
]
|
| 589 |
+
);
|
| 590 |
+
s.addTable(tbl.rows, {
|
| 591 |
+
x: 0.6, y: 2.0, w: 8.8,
|
| 592 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 593 |
+
colW: [1.2, 2.5, 1.0, 4.1],
|
| 594 |
+
rowH: [0.38, 0.35, 0.35, 0.35],
|
| 595 |
+
autoPage: false,
|
| 596 |
+
});
|
| 597 |
+
|
| 598 |
+
// Training status
|
| 599 |
+
s.addText([
|
| 600 |
+
{ text: "Training Status (RegFM + MMD)", options: { bold: true, fontSize: 14, fontFace: FONT_H, color: C.primary, breakLine: true, paraSpaceAfter: 6 } },
|
| 601 |
+
], { x: 0.6, y: 3.4, w: 4.5, h: 0.35 });
|
| 602 |
+
|
| 603 |
+
const statusTbl = styledTable(
|
| 604 |
+
["Step", "L_vel", "L_reg", "L_mmd", "Total"],
|
| 605 |
+
[
|
| 606 |
+
["5k", "0.169", "0.318", "0.025", "0.226"],
|
| 607 |
+
["20k", "0.126", "0.254", "0.017", "0.168"],
|
| 608 |
+
["32k", "0.112", "0.236", "0.016", "0.152"],
|
| 609 |
+
]
|
| 610 |
+
);
|
| 611 |
+
s.addTable(statusTbl.rows, {
|
| 612 |
+
x: 0.6, y: 3.8, w: 5.5,
|
| 613 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 614 |
+
colW: [0.8, 1.0, 1.0, 1.0, 1.0],
|
| 615 |
+
rowH: [0.35, 0.3, 0.3, 0.3],
|
| 616 |
+
autoPage: false,
|
| 617 |
+
});
|
| 618 |
+
|
| 619 |
+
// Key design notes
|
| 620 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 621 |
+
x: 6.5, y: 3.4, w: 3.0, h: 1.8,
|
| 622 |
+
fill: { color: C.offWhite },
|
| 623 |
+
});
|
| 624 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 625 |
+
x: 6.5, y: 3.4, w: 0.07, h: 1.8,
|
| 626 |
+
fill: { color: C.accent },
|
| 627 |
+
});
|
| 628 |
+
s.addText([
|
| 629 |
+
{ text: "Design Notes", options: { bold: true, fontSize: 12, fontFace: FONT_H, color: C.accent, breakLine: true, paraSpaceAfter: 4 } },
|
| 630 |
+
], { x: 6.75, y: 3.5, w: 2.6, h: 0.3 });
|
| 631 |
+
s.addText(bullets([
|
| 632 |
+
"Gate init: alpha ~ 0.05",
|
| 633 |
+
"Diagonal removed from R",
|
| 634 |
+
"Tanh bounds R to [-1, 1]",
|
| 635 |
+
"Backbone: d_model=128",
|
| 636 |
+
], { fontSize: 10.5 }), { x: 6.75, y: 3.85, w: 2.6, h: 1.2 });
|
| 637 |
+
}
|
| 638 |
+
|
| 639 |
+
// ====================================================================
|
| 640 |
+
// SLIDE 11 — Schrodinger Bridge: Approach
|
| 641 |
+
// ====================================================================
|
| 642 |
+
{
|
| 643 |
+
const s = pres.addSlide();
|
| 644 |
+
s.background = { color: C.white };
|
| 645 |
+
addTitle(s, "Schrodinger Bridge: Approach", 11);
|
| 646 |
+
|
| 647 |
+
// Motivation
|
| 648 |
+
s.addText(bullets([
|
| 649 |
+
"FM: noise -> target (unpaired, indirect)",
|
| 650 |
+
"SB: source -> target (optimal transport coupling)",
|
| 651 |
+
"Natural fit for perturbation prediction",
|
| 652 |
+
], { fontSize: 13 }), { x: 0.6, y: 1.15, w: 9.0, h: 1.2 });
|
| 653 |
+
|
| 654 |
+
// Variants table
|
| 655 |
+
const tbl = styledTable(
|
| 656 |
+
["Variant", "Transport", "Score Head", "Anchoring", "Key Feature"],
|
| 657 |
+
[
|
| 658 |
+
["A1 (baseline)", "SB-ODE", "None", "None", "Basic SB formulation"],
|
| 659 |
+
["A5 (full SDE)", "SB-SDE", "Full ASB", "None", "Score + velocity joint"],
|
| 660 |
+
["A6 (DSM aniso)", "SB-SDE", "Anisotropic", "None", "Per-gene noise scale"],
|
| 661 |
+
["SA1 (src-ODE)", "SB-ODE", "None", "Source cell", "Anchor at x_ctrl"],
|
| 662 |
+
["SA6 (src-SDE)", "SB-SDE", "Anisotropic", "Source cell", "Anchor + aniso score"],
|
| 663 |
+
]
|
| 664 |
+
);
|
| 665 |
+
s.addTable(tbl.rows, {
|
| 666 |
+
x: 0.4, y: 2.55, w: 9.2,
|
| 667 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 668 |
+
colW: [1.7, 1.2, 1.4, 1.3, 3.6],
|
| 669 |
+
rowH: [0.38, 0.32, 0.32, 0.32, 0.32, 0.32],
|
| 670 |
+
autoPage: false,
|
| 671 |
+
});
|
| 672 |
+
|
| 673 |
+
// Source-anchored note
|
| 674 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 675 |
+
x: 0.5, y: 4.35, w: 9.0, h: 0.75,
|
| 676 |
+
fill: { color: C.offWhite },
|
| 677 |
+
});
|
| 678 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 679 |
+
x: 0.5, y: 4.35, w: 0.08, h: 0.75,
|
| 680 |
+
fill: { color: C.accent },
|
| 681 |
+
});
|
| 682 |
+
s.addText([
|
| 683 |
+
{ text: "Source-Anchored: ", options: { bold: true, fontSize: 12, color: C.accent } },
|
| 684 |
+
{ text: "ODE starts from x_ctrl (not noise). Loss_v drops to ~0.0004 (vs ~0.3 for standard SB).", options: { fontSize: 12, color: C.bodyText } },
|
| 685 |
+
], { x: 0.8, y: 4.4, w: 8.5, h: 0.6, fontFace: FONT_B });
|
| 686 |
+
}
|
| 687 |
+
|
| 688 |
+
// ====================================================================
|
| 689 |
+
// SLIDE 12 — Schrodinger Bridge: Results
|
| 690 |
+
// ====================================================================
|
| 691 |
+
{
|
| 692 |
+
const s = pres.addSlide();
|
| 693 |
+
s.background = { color: C.white };
|
| 694 |
+
addTitle(s, "Schrodinger Bridge: Results", 12);
|
| 695 |
+
|
| 696 |
+
// Top table — use explicit large font cells
|
| 697 |
+
const hdr12 = ["Model", "Pearson", "MSE", "DE Dir.", "Discrim."].map(h => ({
|
| 698 |
+
text: h, options: { bold: true, color: C.white, fill: { color: C.tableHead }, fontSize: 13, fontFace: FONT_B, align: "center", valign: "middle" },
|
| 699 |
+
}));
|
| 700 |
+
const data12 = [
|
| 701 |
+
[{ text: "scDFM Baseline", bold: true }, { text: "0.866", bold: true }, "0.0032", "0.937", "0.980"],
|
| 702 |
+
[{ text: "SB A1 (baseline)", bold: true }, { text: "0.858", bold: true }, "0.0072", "0.902", "0.957"],
|
| 703 |
+
["SB A6 (aniso DSM)", "0.849", "0.0074", "0.901", "0.956"],
|
| 704 |
+
["SA1 / SA6", "Training...", "@ 195k", "-", "-"],
|
| 705 |
+
].map((row, ri) => row.map(cell => {
|
| 706 |
+
const isObj = typeof cell === "object" && cell !== null;
|
| 707 |
+
return { text: isObj ? cell.text : String(cell), options: { fontSize: 13, fontFace: FONT_B, color: C.darkText, fill: { color: ri % 2 === 0 ? C.white : C.tableAlt }, align: "center", valign: "middle", bold: isObj && cell.bold ? true : false } };
|
| 708 |
+
}));
|
| 709 |
+
s.addTable([hdr12, ...data12], {
|
| 710 |
+
x: 0.5, y: 1.15, w: 9.0,
|
| 711 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 712 |
+
colW: [2.4, 1.6, 1.4, 1.6, 1.6],
|
| 713 |
+
rowH: [0.42, 0.4, 0.4, 0.4, 0.4],
|
| 714 |
+
autoPage: false,
|
| 715 |
+
});
|
| 716 |
+
|
| 717 |
+
// Training loss comparison
|
| 718 |
+
s.addText([
|
| 719 |
+
{ text: "Training Loss Comparison", options: { bold: true, fontSize: 14, fontFace: FONT_H, color: C.primary, breakLine: true, paraSpaceAfter: 6 } },
|
| 720 |
+
], { x: 0.6, y: 3.25, w: 8.8, h: 0.35 });
|
| 721 |
+
|
| 722 |
+
const hdrL = ["Variant", "Loss_v", "Loss_s", "Notes"].map(h => ({
|
| 723 |
+
text: h, options: { bold: true, color: C.white, fill: { color: C.tableHead }, fontSize: 12, fontFace: FONT_B, align: "center", valign: "middle" },
|
| 724 |
+
}));
|
| 725 |
+
const dataL = [
|
| 726 |
+
["A1 Baseline", "0.26 - 0.40", "N/A", "Stable"],
|
| 727 |
+
["A6 DSM Aniso", "0.30 - 0.37", "0.76 - 0.80", "Better score"],
|
| 728 |
+
["SA1 Src-ODE", "~0.0005", "N/A", "Very low (anchored)"],
|
| 729 |
+
["SA6 Src-SDE", "~0.001", "~0.057", "Anchored + aniso"],
|
| 730 |
+
].map((row, ri) => row.map(cell => ({
|
| 731 |
+
text: cell, options: { fontSize: 12, fontFace: FONT_B, color: C.darkText, fill: { color: ri % 2 === 0 ? C.white : C.tableAlt }, align: "center", valign: "middle" },
|
| 732 |
+
})));
|
| 733 |
+
s.addTable([hdrL, ...dataL], {
|
| 734 |
+
x: 0.5, y: 3.65, w: 9.0,
|
| 735 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 736 |
+
colW: [2.2, 2.0, 1.8, 3.0],
|
| 737 |
+
rowH: [0.38, 0.35, 0.35, 0.35, 0.35],
|
| 738 |
+
autoPage: false,
|
| 739 |
+
});
|
| 740 |
+
}
|
| 741 |
+
|
| 742 |
+
// ====================================================================
|
| 743 |
+
// SLIDE 13 — Comprehensive Comparison
|
| 744 |
+
// ====================================================================
|
| 745 |
+
{
|
| 746 |
+
const s = pres.addSlide();
|
| 747 |
+
s.background = { color: C.white };
|
| 748 |
+
addTitle(s, "Comprehensive Comparison", 13);
|
| 749 |
+
|
| 750 |
+
const tbl = styledTable(
|
| 751 |
+
["Method", "Approach", "Pearson", "MSE", "DE Dir.", "Discrim."],
|
| 752 |
+
[
|
| 753 |
+
[{ text: "scDFM Baseline", bold: true }, "Flow Matching", { text: "0.866", bold: true }, { text: "0.003", bold: true }, { text: "0.937", bold: true }, { text: "0.980", bold: true }],
|
| 754 |
+
[{ text: "SB A1", bold: true }, "Schrodinger Bridge", "0.858", "0.007", "0.902", "0.957"],
|
| 755 |
+
["SB A6", "SB + Aniso DSM", "0.849", "0.007", "0.901", "0.956"],
|
| 756 |
+
[{ text: "dim1 ablation", bold: true }, "Cascaded (d=1)", "0.752", "0.006", "0.878", "0.914"],
|
| 757 |
+
["grn_dense4", "Cascaded (d=4)", "0.122", "0.020", "0.780", "0.521"],
|
| 758 |
+
["grn_scalar", "Cascaded (d=1, L2)", "0.087", "0.021", "0.793", "0.534"],
|
| 759 |
+
["grn_att_only", "Cascaded (bilinear)", "-0.097", "0.602", "0.747", "0.552"],
|
| 760 |
+
["grn_svd_cross", "Cascaded (SVD)", "-0.096", "0.575", "0.746", "0.492"],
|
| 761 |
+
["RegFM (20k)", "Structural Bias", "0.040", "0.128", "0.748", "0.505"],
|
| 762 |
+
]
|
| 763 |
+
);
|
| 764 |
+
s.addTable(tbl.rows, {
|
| 765 |
+
x: 0.3, y: 1.1, w: 9.4,
|
| 766 |
+
border: { pt: 0.5, color: "DDDDDD" },
|
| 767 |
+
colW: [1.6, 1.7, 1.1, 1.0, 1.0, 1.0],
|
| 768 |
+
rowH: [0.38, 0.35, 0.35, 0.35, 0.35, 0.35, 0.35, 0.35, 0.35, 0.35],
|
| 769 |
+
autoPage: false,
|
| 770 |
+
});
|
| 771 |
+
|
| 772 |
+
// Color legend
|
| 773 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 774 |
+
x: 0.5, y: 4.85, w: 9.0, h: 0.5,
|
| 775 |
+
fill: { color: C.offWhite },
|
| 776 |
+
});
|
| 777 |
+
s.addText([
|
| 778 |
+
{ text: "Top tier: ", options: { bold: true, fontSize: 11, color: C.green } },
|
| 779 |
+
{ text: "Baseline (0.866), SB A1 (0.858) ", options: { fontSize: 11, color: C.bodyText } },
|
| 780 |
+
{ text: "Mid tier: ", options: { bold: true, fontSize: 11, color: C.accent } },
|
| 781 |
+
{ text: "dim1 (0.752) ", options: { fontSize: 11, color: C.bodyText } },
|
| 782 |
+
{ text: "Failed: ", options: { bold: true, fontSize: 11, color: C.red } },
|
| 783 |
+
{ text: "All high-dim cascaded variants", options: { fontSize: 11, color: C.bodyText } },
|
| 784 |
+
], { x: 0.7, y: 4.85, w: 8.6, h: 0.5, fontFace: FONT_B, valign: "middle" });
|
| 785 |
+
}
|
| 786 |
+
|
| 787 |
+
// ====================================================================
|
| 788 |
+
// SLIDE 14 — Key Takeaways
|
| 789 |
+
// ====================================================================
|
| 790 |
+
{
|
| 791 |
+
const s = pres.addSlide();
|
| 792 |
+
s.background = { color: C.white };
|
| 793 |
+
addTitle(s, "Key Takeaways", 14);
|
| 794 |
+
|
| 795 |
+
const takeaways = [
|
| 796 |
+
{ num: "1", title: "Cascaded GRN Generation Fails", desc: "High-dim latent target (G x G sparse) is fundamentally too hard to generate via flow matching." },
|
| 797 |
+
{ num: "2", title: "SB Competitive with FM Baseline", desc: "Schrodinger Bridge (A1: 0.858) nearly matches scDFM (0.866); source-anchored variants training." },
|
| 798 |
+
{ num: "3", title: "RegFM: New Paradigm", desc: "Treat delta_attn as privileged info; embed GRN as structural bias, not generation target." },
|
| 799 |
+
{ num: "4", title: "dim1 Confirms the Diagnosis", desc: "Only 1d latent converges; validates that task difficulty scales with latent dimensionality." },
|
| 800 |
+
];
|
| 801 |
+
|
| 802 |
+
takeaways.forEach((t, i) => {
|
| 803 |
+
const ty = 1.15 + i * 1.05;
|
| 804 |
+
// Number circle
|
| 805 |
+
s.addShape(pres.shapes.OVAL, {
|
| 806 |
+
x: 0.6, y: ty + 0.05, w: 0.45, h: 0.45,
|
| 807 |
+
fill: { color: C.primary },
|
| 808 |
+
});
|
| 809 |
+
s.addText(t.num, {
|
| 810 |
+
x: 0.6, y: ty + 0.05, w: 0.45, h: 0.45,
|
| 811 |
+
fontSize: 16, fontFace: FONT_H, color: C.white, bold: true, align: "center", valign: "middle",
|
| 812 |
+
});
|
| 813 |
+
// Title + description
|
| 814 |
+
s.addText(t.title, {
|
| 815 |
+
x: 1.25, y: ty, w: 8.2, h: 0.3,
|
| 816 |
+
fontSize: 15, fontFace: FONT_H, color: C.primary, bold: true, margin: 0,
|
| 817 |
+
});
|
| 818 |
+
s.addText(t.desc, {
|
| 819 |
+
x: 1.25, y: ty + 0.32, w: 8.2, h: 0.55,
|
| 820 |
+
fontSize: 12, fontFace: FONT_B, color: C.bodyText, margin: 0,
|
| 821 |
+
});
|
| 822 |
+
});
|
| 823 |
+
}
|
| 824 |
+
|
| 825 |
+
// ====================================================================
|
| 826 |
+
// SLIDE 15 — Next Steps
|
| 827 |
+
// ====================================================================
|
| 828 |
+
{
|
| 829 |
+
const s = pres.addSlide();
|
| 830 |
+
s.background = { color: C.dark };
|
| 831 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 832 |
+
x: 0, y: 0, w: 0.12, h: 5.625, fill: { color: C.primary },
|
| 833 |
+
});
|
| 834 |
+
|
| 835 |
+
s.addText("Next Steps", {
|
| 836 |
+
x: 0.7, y: 0.6, w: 8.6, h: 0.6,
|
| 837 |
+
fontSize: 32, fontFace: FONT_H, color: C.white, bold: true, margin: 0,
|
| 838 |
+
});
|
| 839 |
+
|
| 840 |
+
const steps = [
|
| 841 |
+
{ title: "RegFM Full Training", desc: "Continue to 100k+ steps; evaluate and compare with baseline at convergence" },
|
| 842 |
+
{ title: "SB Source-Anchored Eval", desc: "Evaluate SA1 / SA6 at 200k; compare ODE vs SDE transport" },
|
| 843 |
+
{ title: "RegFM vs SB vs Baseline", desc: "Head-to-head comparison on cell-eval + distributional metrics" },
|
| 844 |
+
{ title: "Distributional Evaluation", desc: "Apply new metrics (MMD, Energy Distance, C2ST, kNN) beyond conditional mean" },
|
| 845 |
+
{ title: "Interpretability Analysis", desc: "Visualize R_theta from RegFM; compare with known GRN structure" },
|
| 846 |
+
];
|
| 847 |
+
|
| 848 |
+
steps.forEach((st, i) => {
|
| 849 |
+
const sy = 1.5 + i * 0.78;
|
| 850 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 851 |
+
x: 0.7, y: sy, w: 8.6, h: 0.65,
|
| 852 |
+
fill: { color: "2A2A42" },
|
| 853 |
+
});
|
| 854 |
+
s.addShape(pres.shapes.RECTANGLE, {
|
| 855 |
+
x: 0.7, y: sy, w: 0.07, h: 0.65,
|
| 856 |
+
fill: { color: C.primary },
|
| 857 |
+
});
|
| 858 |
+
s.addText(st.title, {
|
| 859 |
+
x: 1.0, y: sy + 0.03, w: 3.0, h: 0.3,
|
| 860 |
+
fontSize: 14, fontFace: FONT_B, color: C.white, bold: true, margin: 0,
|
| 861 |
+
});
|
| 862 |
+
s.addText(st.desc, {
|
| 863 |
+
x: 1.0, y: sy + 0.32, w: 8.1, h: 0.28,
|
| 864 |
+
fontSize: 12, fontFace: FONT_B, color: "CCCCCC", margin: 0,
|
| 865 |
+
});
|
| 866 |
+
});
|
| 867 |
+
|
| 868 |
+
addSlideNum(s, 15);
|
| 869 |
+
}
|
| 870 |
+
|
| 871 |
+
// ── Write file ─────────────────────────────────────────────────────
|
| 872 |
+
pres.writeFile({ fileName: "/home/hp250092/ku50001222/qian/aivc/lfj/Report/week10/GRN_Progress_Report.pptx" })
|
| 873 |
+
.then(() => console.log("PPTX saved successfully."))
|
| 874 |
+
.catch((err) => console.error("Error:", err));
|