Add evaluated crossdocked-scaffold checkpoint and metrics; simplify GEOM scaffold name

README.md

@@ -8,6 +8,8 @@ tags:
   - qm9
   - geom-drugs
   - chembl3d
+  - crossdocked
+  - structure-based-drug-design
   - 3d
   - arxiv:2606.07239
 ---
@@ -15,8 +17,8 @@ tags:
 # Morph: 3D molecular generation and scaffold decoration
 
 Checkpoints for **unconditional 3D molecular generation** on **QM9**,
-**GEOM-Drugs** and **ChEMBL3D**, plus **GEOM-Drugs scaffold decoration** with
-rank-16 LoRA after 25 fine-tuning epochs. The releases include the configurations,
+**GEOM-Drugs** and **ChEMBL3D**, plus **GEOM-Drugs scaffold decoration**
+and **CrossDocked scaffold decoration in protein pockets**. The releases include the configurations,
 vocabularies and training distributions needed for their inference workflows.
 Unconditional generation requires no training dataset; scaffold decoration uses
 a supplied scaffold and the dedicated [scaffold bundle](#scaffold-decoration).
@@ -45,14 +47,14 @@ New catalog entries are discovered without updating the Python package.
 ## Scaffold decoration
 
 **GEOM-Drugs scaffold decoration is also available:** the
-[`geom-scaffold-murcko-substituent-lora-r16` bundle](https://huggingface.co/maltefranke/morph/tree/main/geom/scaffold_murcko_substituent_lora_r16)
+[`geom-scaffold-murcko` bundle](https://huggingface.co/maltefranke/morph/tree/main/geom/scaffold_murcko_substituent_lora_r16)
 contains the checkpoint after **25 fine-tuning epochs**, initialized from the
 GEOM explicit-H epoch-100 EMA weights. It uses **rank-16 LoRA**, trainable output
 heads and substituent-constrained insertion supervision.
 
 | Model | Hydrogen representation | Completed fine-tuning epochs | Checkpoint |
 | --- | --- | ---: | --- |
-| `geom-scaffold-murcko-substituent-lora-r16` | Explicit H atoms | 25 | [epoch_025.ckpt](https://huggingface.co/maltefranke/morph/resolve/main/geom/scaffold_murcko_substituent_lora_r16/epoch_025.ckpt) |
+| `geom-scaffold-murcko` | Explicit H atoms | 25 | [epoch_025.ckpt](https://huggingface.co/maltefranke/morph/resolve/main/geom/scaffold_murcko_substituent_lora_r16/epoch_025.ckpt) |
 
 The model generates decorations conditioned on a Murcko scaffold. Scaffold atom
 identities, internal bonds and charges are preserved; scaffold positions can move.
@@ -64,11 +66,29 @@ The bundle includes the original checkpoint, training configuration and source,
 portable configuration, vocabularies, distributions, evaluation runtime and fixed
 validation/test scaffold panels. See its [usage and reproduction instructions](https://huggingface.co/maltefranke/morph/blob/main/geom/scaffold_murcko_substituent_lora_r16/README.md).
 
+
+### Scaffold decoration in a protein pocket
+
+[`crossdocked-scaffold`](https://huggingface.co/maltefranke/morph/tree/main/crossdocked/scaffold)
+is the evaluated CrossDocked checkpoint after **100 fine-tuning epochs**. It uses
+implicit ligand hydrogens, rank-8 LoRA and 10% in-vacuum ChEMBL3D replay.
+
+| Model | Hydrogen representation | Completed fine-tuning epochs | Checkpoint |
+| --- | --- | ---: | --- |
+| `crossdocked-scaffold` | Heavy atoms; implicit H | 100 | [epoch_100.ckpt](https://huggingface.co/maltefranke/morph/resolve/main/crossdocked/scaffold/epoch_100.ckpt) |
+
+The bundle contains the evaluated checkpoint, portable configuration,
+vocabulary/statistics assets and complete metrics table.
+Use the [pocket evaluation workflow](https://huggingface.co/maltefranke/morph/blob/main/crossdocked/scaffold/README.md)
+with a scaffold and protein pocket. Like the GEOM scaffold model, it is separate
+from the unconditional CLI/catalog.
+
 ## Evaluation results
 
 Values are **mean ± sample standard deviation** across seeds **42, 43 and 44**
-(`ddof=1`). Rates are percentages; diversity is on a 0–1 scale.
-Both evaluations use EMA weights and 100 integration steps.
+(`ddof=1`). Rates are percentages; diversity and similarity are on a 0–1 scale.
+Vina scores are in kcal/mol.
+All evaluations use EMA weights and 100 integration steps.
 
 <details>
 <summary><strong>Unconditional generation — all 10 checkpoints</strong></summary>
@@ -104,7 +124,7 @@ The generated coordinates are not relaxed.
 
 | Model | Completed epochs | Novelty (%) | PoseBusters validity (%) | Diversity (0–1) | Uniqueness (%) | Connected-sanitization validity (%) |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
-| GEOM-Drugs · scaffold decoration (LoRA r16) | 25 (fine-tuning) | 97.52 ± 0.10 | 94.10 ± 0.19 | 0.9076 ± 0.0004 | 63.54 ± 0.59 | 97.99 ± 0.20 |
+| `geom-scaffold-murcko` | 25 (fine-tuning) | 97.52 ± 0.10 | 94.10 ± 0.19 | 0.9076 ± 0.0004 | 63.54 ± 0.59 | 97.99 ± 0.20 |
 
 **PoseBusters protocol:** the native scaffold evaluation checks chemistry and distance
 geometry, excluding the full offline energy and flatness checks. The value above is
@@ -119,7 +139,31 @@ for the saved evaluation runtime and test panel.
 
 </details>
 
-Metric definitions:
+<details>
+<summary><strong>Scaffold decoration in a pocket — CrossDocked, 100 fine-tuning epochs</strong></summary>
+
+`crossdocked-scaffold`, evaluated on the published DiffDec single- and multi-decoration
+test panels, with **100 attempts per condition per seed** (43,500 total).
+
+| Panel | Conditions | Attempts per seed | Validity (%) | Uniqueness (%) | Similarity (0–1) | Recovery (%) | Vina score (kcal/mol) | High affinity (%) |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| Single | 43 | 4300 | 88.72 ± 0.71 | 53.32 ± 1.02 | 0.8157 ± 0.0042 | 39.00 ± 1.15 | -8.041 ± 0.007 | 44.18 ± 3.67 |
+| Multi | 102 | 10200 | 89.80 ± 0.16 | 66.43 ± 0.38 | 0.7671 ± 0.0032 | 37.83 ± 5.33 | -8.088 ± 0.010 | 40.33 ± 3.14 |
+
+Mean ± sample standard deviation across seeds 42/43/44. The six columns use
+DiffDec's grouping and formulas; validity and uniqueness have different definitions
+from the unconditional and GEOM scaffold tables. Docking re-embeds and UFF-minimizes
+prediction SMILES, so it does not score the original generated coordinates.
+Invalid predictions are replaced with the scaffold for docking. Reproduction
+differences include reconstructed reference grouping strings, predicted bond
+topology, newer RDKit and deterministic embedding seeds.
+
+See the [full CSV, including individual seeds and docking failures](https://huggingface.co/maltefranke/morph/blob/main/crossdocked/scaffold/summary.csv)
+and [protocol and reproduction notes](https://huggingface.co/maltefranke/morph/blob/main/crossdocked/scaffold/README.md).
+
+</details>
+
+Metric definitions for the unconditional and GEOM scaffold tables:
 
 - **Novelty:** fraction of generated, SMILES-convertible molecules absent from the
   corresponding training reference, retaining duplicates in the denominator.

crossdocked/scaffold/README.md

@@ -0,0 +1,60 @@
+# crossdocked-scaffold
+
+Morph scaffold decoration conditioned on a protein pocket, evaluated after
+**100 completed fine-tuning epochs (40,000 steps)**. The model uses implicit ligand
+hydrogens, rank-8 LoRA and trainable output heads, with 90% CrossDocked pocket
+examples and 10% in-vacuum ChEMBL3D replay. Scaffold atom identities, internal
+bonds and charges are preserved; scaffold positions can move.
+
+## Evaluation
+
+**Mean ± sample standard deviation (ddof=1), seeds 42, 43 and 44**, EMA weights,
+100 integration steps, float32 sampling. Each published DiffDec condition receives
+100 generation attempts: **43,500 attempts in total**.
+
+| Panel | Conditions | Attempts per seed | Validity (%) | Uniqueness (%) | Similarity (0–1) | Recovery (%) | Vina score (kcal/mol) | High affinity (%) |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| Single | 43 | 4300 | 88.72 ± 0.71 | 53.32 ± 1.02 | 0.8157 ± 0.0042 | 39.00 ± 1.15 | -8.041 ± 0.007 | 44.18 ± 3.67 |
+| Multi | 102 | 10200 | 89.80 ± 0.16 | 66.43 ± 0.38 | 0.7671 ± 0.0032 | 37.83 ± 5.33 | -8.088 ± 0.010 | 40.33 ± 3.14 |
+
+The [complete metrics CSV](summary.csv) includes all six individual seed/panel
+results, means, standard deviations, chemistry validity diagnostics, reference
+docking affinities and docking failure counts. High affinity is shown as a
+percentage here and stored as a fraction in the CSV.
+
+Validity, uniqueness, similarity and recovery use DiffDec's grouping and formulas.
+These validity and uniqueness definitions differ from the unconditional and GEOM
+scaffold chemistry metrics. Duplicate published conditions remain in the panels.
+
+Docking uses QuickVina 2, a 30 Å box, exhaustiveness 16 and docking seed 1.
+Predicted SMILES are re-embedded and UFF-minimized before docking; these scores
+do not assess the original generated coordinates. Invalid predictions are replaced
+with the scaffold, following the released DiffDec convention. Across the three
+seeds, prediction docking failures total 55 for the single panel and 334 for the
+multi panel; all 435 reference docking tasks succeeded, with no timeouts.
+
+Reproduction differences from upstream DiffDec include reconstructed reference
+grouping strings because the original CSVs are unavailable, Morph's predicted
+bond topology, RDKit 2025.9.6 and deterministic per-attempt embedding seeds.
+
+## Model files
+
+- `epoch_100.ckpt`: the exact evaluated Lightning checkpoint, including current
+  weights, EMA weights and optimizer state. Its SHA-256 is recorded in `metadata.json`.
+- `config.yaml`: the saved architecture and sampling configuration with portable
+  data/asset paths and base-checkpoint loading disabled.
+- Ordered atom/edge/charge vocabularies and `distributions.pt`: matching model assets.
+- `metadata.json`: model identity, evaluation settings and vocabulary metadata.
+- `summary.csv`: the full metrics table; `manifest.json`: file sizes and SHA-256 hashes.
+
+```bash
+hf download maltefranke/morph --include 'crossdocked/scaffold/*' --local-dir morph-models
+export MORPH_BUNDLE_DIR="$PWD/morph-models/crossdocked/scaffold"
+```
+
+Use Morph's scaffold-pocket evaluation workflow with this checkpoint and
+`config.yaml`, passing the bundle folder as `--assets-root`. Sampling requires
+a scaffold and protein pocket; the unconditional `morph generate` interface and
+catalog do not load this conditional model. The matching processed data are in
+[morph-data](https://huggingface.co/datasets/maltefranke/morph-data/tree/main/crossdocked),
+and the published test tensors are in [DiffDec's Zenodo release](https://zenodo.org/records/10527451).

crossdocked/scaffold/atom_tokens.txt

@@ -0,0 +1,12 @@
+B
+Br
+C
+Cl
+F
+I
+N
+O
+P
+S
+Se
+Si

crossdocked/scaffold/charge_tokens.txt

@@ -0,0 +1,3 @@
+-1
+0
+1

crossdocked/scaffold/config.yaml

@@ -0,0 +1,474 @@
+representation: molecule
+seed: 42
+heads:
+  heads:
+    _target_: morph.model.layers.heads.MultiHeadModule
+    output_init_std: ${model.head_initialization.output_std}
+    independent_adapters: ${oc.select:model.head_independent_adapters,false}
+    heads_configs:
+      graph_aggregation: mean
+      node_heads:
+        atom_type_head:
+          input_dim: ${model.head_node_dim}
+          output_dim: ${len:${data.vocab.atom_tokens}}
+          adapter_dim: ${model.head_node_adapter_dim}
+          group: state
+        charge_head:
+          input_dim: ${model.head_node_dim}
+          output_dim: ${len:${data.vocab.charge_tokens}}
+          adapter_dim: ${model.head_node_adapter_dim}
+          group: state
+        ins_rate_head:
+          input_dim: ${model.head_node_dim}
+          output_dim: 1
+          adapter_dim: ${model.head_node_adapter_dim}
+          group: edit
+          initial_rate: ${model.head_initialization.ins_rate}
+        do_del_head:
+          input_dim: ${model.head_node_dim}
+          output_dim: 1
+          adapter_dim: ${model.head_node_adapter_dim}
+          group: edit
+          initial_probability: ${model.head_initialization.do_del_probability}
+        do_sub_a_head:
+          input_dim: ${model.head_node_dim}
+          output_dim: 1
+          adapter_dim: ${model.head_node_adapter_dim}
+          group: edit
+          initial_probability: ${model.head_initialization.do_sub_a_probability}
+      edge_heads:
+        edge_type_head:
+          input_dim: ${model.head_edge_dim}
+          output_dim: ${len:${data.vocab.edge_tokens}}
+          adapter_dim: ${model.head_edge_adapter_dim}
+          group: state
+        do_sub_e_head:
+          input_dim: ${model.head_edge_dim}
+          output_dim: 1
+          adapter_dim: ${model.head_edge_adapter_dim}
+          group: state
+          initial_probability: ${model.head_initialization.do_sub_e_probability}
+  ins_edge_head:
+    _target_: morph.model.layers.heads.InsertionEdgeHead
+    _recursive_: false
+    hidden_dim: ${model.head_node_dim}
+    readout_dim: ${model.head_insertion_edge_dim}
+    output_init_std: ${model.head_initialization.output_std}
+    n_edge_types: ${len:${data.vocab.edge_tokens}}
+    n_atom_types: ${len:${data.vocab.atom_tokens}}
+    n_charge_types: ${len:${data.vocab.charge_tokens}}
+    rbf_embedding_args:
+      _target_: morph.model.layers.embedding.RBFEmbedding
+      num_rbf: 128
+      rbf_dmax: 3.5
+      out_dim: 64
+      dropout: 0.0
+      trainable: false
+insertion_head:
+  mode: unimodal
+  dimension: 3
+  num_components: 1
+  model:
+    _target_: morph.model.layers.heads.UnimodalInsertionHead
+    hidden_dim: ${model.head_node_dim}
+    readout_dim: ${model.head_insertion_adapter_dim}
+    output_init_std: ${model.head_initialization.output_std}
+    independent_adapters: ${oc.select:model.head_independent_adapters,false}
+    N_a: ${len:${data.vocab.atom_tokens}}
+    N_c: ${len:${data.vocab.charge_tokens}}
+model:
+  module:
+    optimizer_config:
+      implementation: torch
+      optimizer:
+        execution: ${oc.env:MORPH_MUON_EXECUTION,native}
+        lr: 0.001
+        momentum: 0.95
+        weight_decay: 0.0
+        nesterov: true
+        ns_steps: 5
+        eps: 1.0e-07
+        adjust_lr_fn: original
+      aux_optimizer:
+        lr: 0.0001
+        betas:
+        - 0.9
+        - 0.95
+        eps: 1.0e-10
+        weight_decay: 0.0
+      gradient_clip_val: 1.0
+      gradient_clip_algorithm: norm
+      scheduler:
+        _target_: morph.utils.lr_schedulers.MultiGroupCosineWarmupLR
+        warmup_steps: 1000
+        total_steps: 40000
+        min_lr_fraction: 0.05
+        last_epoch: -1
+      interval: step
+      monitor: null
+    loss_weights:
+      do_sub_a: 0.2
+      sub_a_class: 0.03
+      do_sub_e: 0.25
+      sub_e_class: 0.5
+      do_del: 0.4
+      ins_rate: 0.15
+      ins_state: 0.01
+      ins_e: 0.4
+      x: 2.0
+      c: 0.5
+      ins_e_ii: 0.03
+    _target_: morph.model.finetune.scaffold.LightningModuleRatesScaffoldFinetune
+    insertion_head_mode: ${insertion_head.mode}
+    ins_noise_scale: ${data.ins_noise_scale}
+    ema_decay: 0.999
+    ema_warmup_steps: 5000
+    use_ema_for_eval: true
+    time_dist: ${if:${eq:${model.sample_time_strategy}, log}, ${model.beta_sched},
+      ${model.linear_sched}}
+    integrator: ${model.integrator}
+    vocab: ${data.vocab}
+    use_learnable_loss_weights: false
+    use_time_weights: false
+    model:
+      _target_: morph.model.finetune.pocket.ScaffoldPocketConditionalDiTBackboneWithHeads
+      _recursive_: false
+      atom_type_embedding_args: ${embeddings.atom_type_embedding}
+      edge_type_embedding_args: ${embeddings.edge_type_embedding}
+      time_embedding_args: ${embeddings.time_embedding}
+      node_count_embedding_args: ${embeddings.node_count_embedding}
+      incident_bond_embedding: false
+      backbone_model_args: ${model.backbone}
+      heads_args: ${heads.heads}
+      insertion_head_args: ${insertion_head.model}
+      ins_edge_head_args: ${heads.ins_edge_head}
+      encoder_args: ${model.pocket_encoder}
+      use_distance_bias: true
+    pretrained_ckpt: null
+    pretrained_prefer_ema: true
+    freeze_pretrained_backbone: true
+    freeze_output_heads: false
+    exclude_scaffold_from_sub_loss: true
+    exclude_scaffold_from_x_loss: false
+    freeze_keep_positions: false
+    freeze_keep_charges: true
+  rotate: true
+  n_augmentations: 1
+  integrator: ${integrator.integrator}
+  beta_sched:
+    _target_: torch.distributions.Beta
+    concentration1: 2.0
+    concentration0: 1.0
+  linear_sched:
+    _target_: torch.distributions.Uniform
+    low: 0.0
+    high: 1.0
+  sample_time_strategy: linear
+  head_node_dim: ${model.backbone.out_node_nf}
+  head_edge_dim: ${model.backbone.out_edge_nf}
+  head_node_adapter_dim: 256
+  head_edge_adapter_dim: 256
+  head_insertion_adapter_dim: 256
+  head_insertion_edge_dim: 256
+  head_independent_adapters: true
+  head_initialization:
+    output_std: 0.001
+    coordinate_output_std: 0.0
+    do_del_probability: 0.5
+    do_sub_a_probability: 0.5
+    do_sub_e_probability: 0.5
+    ins_rate: 0.6931471805599453
+  token_weighting:
+    type_loss_token_weights: uniform
+    weight_alpha: 0.4
+  lora:
+    enabled: true
+    rank: 8
+    alpha: 8.0
+    dropout: 0.0
+    targets:
+    - qkv
+    - proj
+    - mlp.gate_value
+    - mlp.out_proj
+    freeze_base: true
+  backbone:
+    _target_: morph.model.backbones.conditional_dit.ConditionalDiTBackbone
+    _recursive_: false
+    d_input: ${embeddings.atom_type_embedding.out_dim}
+    d_cond: ${oc.eval:"${embeddings.time_embedding.out_dim} + ${embeddings.node_count_embedding.out_dim}"}
+    d_model: 512
+    out_node_nf: 256
+    out_edge_nf: 256
+    num_layers: 8
+    nhead: 16
+    mlp_ratio: 4.0
+    ffn_hidden_dim: null
+    ffn_multiple_of: 64
+    qk_norm: true
+    coordinate_output_init_std: ${model.head_initialization.coordinate_output_std}
+    center_coordinate_output: true
+    in_edge_nf: ${embeddings.edge_type_embedding.out_dim}
+    edge_pair_dim: 256
+    pair_bias_hidden_dim: 32
+    num_atom_types: ${len:${data.vocab.atom_tokens}}
+    lora: ${model.lora}
+    rbf_embedding_args:
+      _target_: morph.model.layers.embedding.RBFEmbedding
+      num_rbf: 128
+      rbf_dmax: 3.5
+      out_dim: 64
+      dropout: 0.0
+      trainable: false
+    ctx_dim: 128
+    use_cross_attn: true
+    use_node_cond: true
+    use_edge_cond: true
+  pocket_encoder:
+    _target_: morph.model.finetune.pocket.PocketContextEncoder
+    num_atom_types: ${len:${data.pocket_vocab.atom_tokens}}
+    num_charges: ${len:${data.pocket_vocab.charge_tokens}}
+    ctx_dim: ${model.backbone.ctx_dim}
+    num_attn_layers: 2
+    nhead: 8
+    dropout: 0.0
+schedule:
+  _target_: morph.flow_matching.schedules.SmoothstepSchedule
+  shift: 0.8
+trainer:
+  trainer:
+    deterministic: warn
+    devices: 2
+    accelerator: gpu
+    strategy: ddp_find_unused_parameters_false
+    precision: bf16-mixed
+    max_epochs: 100
+    accumulate_grad_batches: 4
+    num_sanity_val_steps: 0
+    gradient_clip_val: null
+    gradient_clip_algorithm: norm
+    check_val_every_n_epoch: 5
+    num_val_samples: null
+    limit_val_batches: null
+    limit_predict_batches: 4
+    use_distributed_sampler: false
+    enable_progress_bar: false
+    max_time:
+      seconds: 84526
+  do_train: true
+  do_validate: false
+  do_predict: false
+  checkpoint_path: null
+  final_checkpoint_path: final.ckpt
+callbacks:
+  monitor_metric: ${if:${oc.eval:"'${representation}' in ('geometric_graph', 'molecule')"},val/validity,val/batch/pair_dist_l1}
+  monitor_metric_mode: ${if:${oc.eval:"'${representation}' in ('geometric_graph',
+    'molecule')"},max,min}
+  model_checkpoints:
+    save_top_k: 1
+    save_last: false
+    verbose: false
+    every_n_epochs: 5
+  every_n_epochs_checkpoint:
+    every_n_epochs: 5
+    save_top_k: -1
+    save_last: link
+    verbose: false
+    save_on_train_epoch_end: true
+  metric_audit:
+    _target_: run_audit.ValidationAudit
+    expected_samples: 4096
+    expected_scaffolds: 512
+logging:
+  enabled: false
+  project: morph-crossdocked-scaffold-finetune
+  name: ${oc.env:MORPH_RUN_ID,base}
+  save_dir: .
+  offline: false
+  tags:
+  - pocket-scaffold-decoration
+  - chembl3d-no-h
+  - libinvent
+  - lora
+  - trainable-heads
+  - movable-scaffold
+  - jitter-0.1A
+  - replay-10pct
+  - scaffold-metrics
+integrator:
+  integrator:
+    _target_: morph.flow_matching.integration.RateIntegrator
+    vocab: ${data.vocab}
+    distributions: None
+    insertion_head_mode: ${insertion_head.mode}
+    insertion_dimension: ${insertion_head.dimension}
+    insertion_num_components: ${insertion_head.num_components}
+    time_strategy: log
+    time_beta_alpha: 2.0
+    time_beta_beta: 1.0
+    num_integration_steps: 100
+    ins_noise_scale: ${data.ins_noise_scale}
+    renoise_insertions: true
+    exact_schedule_integration: true
+    del_schedule: ${data.interpolator.del_schedule}
+    ins_schedule: ${data.interpolator.ins_schedule}
+    sub_schedule: ${data.interpolator.sub_schedule}
+    sub_e_schedule: ${data.interpolator.sub_e_schedule}
+data:
+  root_path: ${oc.env:MORPH_DATA_ROOT,./data}
+  preprocess_workers: 10
+  preprocess_chunk_size: 128
+  statistics_workers: ${data.preprocess_workers}
+  hydrogen_mode: implicit
+  cache_name: ${data.dataset_name}${if:${eq:${data.hydrogen_mode},implicit},_implicit_h,''}
+  source_dir: ${oc.env:MORPH_CROSSDOCKED_RAW,null}
+  max_atoms: null
+  allowed_atom_symbols: null
+  n_atoms_strategy: uniform
+  uniform_n_atoms_padding: 5
+  optimal_transport: equivariant
+  ins_noise_scale: 0.65
+  move_noise_scale: 0.5
+  allow_charged: true
+  preprocessing:
+    _target_: morph.dataset.core.preprocessing.Preprocessing
+    _recursive_: false
+    root: ${data.pretrained_assets_root}
+    num_workers: ${data.statistics_workers}
+    train_dataset: null
+    atom_tokens_path: ${data.preprocessing.root}/atom_tokens.txt
+    edge_tokens_path: ${data.preprocessing.root}/edge_tokens.txt
+    charge_tokens_path: ${data.preprocessing.root}/charge_tokens.txt
+    distributions_path: ${data.preprocessing.root}/distributions.pt
+  vocab: null
+  interpolator:
+    _target_: morph.flow_matching.interpolation.Interpolator
+    distributions: null
+    optimal_transport: ${data.optimal_transport}
+    ins_noise_scale: ${data.ins_noise_scale}
+    move_noise_scale: ${data.move_noise_scale}
+    atom_type_noise_prob: 0.0
+    pre_align_iters: 2
+    c_move: 1.0
+    c_sub: 0.0
+    c_ins: 6.125
+    c_del: 6.125
+    del_schedule: ${schedule}
+    ins_schedule: ${schedule}
+    sub_schedule:
+      _target_: morph.flow_matching.schedules.SmoothstepSchedule
+      shift: 1.5
+    sub_e_schedule:
+      _target_: morph.flow_matching.schedules.SmoothstepSchedule
+      shift: 1.5
+    jitter_scale: ${data.jitter_scale}
+  datamodule:
+    _target_: morph.dataset.conditional.pocket.scaffold_pocket.ScaffoldPocketDataModule
+    vocab: ${data.vocab}
+    distributions: null
+    representation: ${representation}
+    datasets:
+      train:
+        _target_: ${data.dataset_target}
+        root: ${data.root}
+        preprocess_workers: ${data.preprocess_workers}
+        preprocess_chunk_size: ${data.preprocess_chunk_size}
+        hydrogen_mode: ${data.hydrogen_mode}
+        source_dir: ${data.source_dir}
+        max_atoms: ${data.max_atoms}
+        allowed_atom_symbols: null
+        vocab: ${data.vocab}
+        distributions: null
+        max_pocket_atoms: ${data.max_pocket_atoms}
+        val_fraction: ${data.val_fraction}
+      val:
+      - _target_: ${data.dataset_target}
+        root: ${data.root}
+        source_dir: ${data.source_dir}
+        hydrogen_mode: ${data.hydrogen_mode}
+        max_pocket_atoms: ${data.max_pocket_atoms}
+        val_fraction: ${data.val_fraction}
+        preprocess_workers: ${data.preprocess_workers}
+      test:
+      - _target_: ${data.dataset_target}
+        root: ${data.root}
+        source_dir: ${data.source_dir}
+        hydrogen_mode: ${data.hydrogen_mode}
+        max_pocket_atoms: ${data.max_pocket_atoms}
+        val_fraction: ${data.val_fraction}
+        preprocess_workers: ${data.preprocess_workers}
+    interpolator: ${data.interpolator}
+    rotate: true
+    n_augmentations: ${model.n_augmentations}
+    num_workers:
+      train: 6
+      val: 2
+      test: 2
+    batch_size:
+      train: 32
+      val: 8
+      test: 8
+    n_atoms_strategy: ${data.n_atoms_strategy}
+    uniform_n_atoms_padding: ${data.uniform_n_atoms_padding}
+    time_dist: ${model.module.time_dist}
+    pocket_vocab: ${data.pocket_vocab}
+    finetune_fraction: ${data.finetune_fraction}
+    train_epoch_size: ${data.train_epoch_size}
+    mix_seed: ${seed}
+    pretrain_dataset:
+      _target_: morph.dataset.unconditional.chembl3d.FlowMatchingChEMBL3DDataset
+      root: ${data.pretrain_root}
+      source_dir: ${data.pretrain_source_dir}
+      hydrogen_mode: ${data.hydrogen_mode}
+      max_atoms: ${data.pretrain_max_atoms}
+      allowed_atom_symbols: ${data.pretrain_allowed_atom_symbols}
+      preprocess_workers: ${data.preprocess_workers}
+      preprocess_chunk_size: ${data.preprocess_chunk_size}
+    scaffold_groups_dir: ${data.scaffold_groups_dir}
+    scaffold_kind: ${data.scaffold_kind}
+    enumerate_slicings: ${data.enumerate_slicings}
+    eval_panel: ${data.eval_panel}
+    train_prefetch_factor: 1
+  dataset_name: crossdocked
+  dataset_target: morph.dataset.conditional.pocket.CrossDockedPocketDataset
+  root: ${oc.env:MORPH_CROSSDOCKED_ROOT,./data/crossdocked}
+  pretrained_assets_root: ${oc.env:MORPH_BUNDLE_DIR,.}
+  pretrain_root: ${data.root_path}/chembl3d_max100_implicit_h
+  pretrain_source_dir: ${oc.env:CHEMBL3D_SOURCE_DIR,null}
+  pretrain_max_atoms: 100
+  pretrain_allowed_atom_symbols: null
+  finetune_fraction: 0.9
+  train_epoch_size: 102400
+  pocket_vocab: null
+  max_pocket_atoms: 512
+  val_fraction: 0.02
+  scaffold_kind: libinvent
+  scaffold_groups_dir: ${data.root}/scaffold_groups
+  enumerate_slicings: true
+  eval_panel:
+    n_scaffolds: 512
+    samples_per_scaffold: 8
+    seed: 42
+  jitter_scale: 0.1
+embeddings:
+  atom_type_embedding:
+    _target_: morph.model.layers.embedding.Embedding
+    num_embeddings: ${len:${data.vocab.atom_tokens}}
+    embedding_dim: 64
+    out_dim: 128
+  edge_type_embedding:
+    _target_: morph.model.layers.embedding.Embedding
+    num_embeddings: ${len:${data.vocab.edge_tokens}}
+    embedding_dim: 64
+    out_dim: 128
+  time_embedding:
+    _target_: morph.model.layers.embedding.TimeEmbedding
+    embedding_dim: 128
+    out_dim: 128
+  node_count_embedding:
+    _target_: morph.model.layers.embedding.ExtrapolatableCountEmbedding
+    embedding_dim: 64
+    out_dim: 128
+    min_period: 128.0
+    max_period: 1024.0
+    max_count: 96.0

crossdocked/scaffold/distributions.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba4101ee7f3284dd2087b15573f1730aaa70ee808d874ced396fcd64d55a66d5
+size 89925

crossdocked/scaffold/edge_tokens.txt

@@ -0,0 +1,4 @@
+<NO_BOND>
+1
+2
+3

crossdocked/scaffold/epoch_100.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d17db9a77e75e12c452f0512301e5894b2d1a0bd9725186e35bf183a238b2b6
+size 625018500

crossdocked/scaffold/manifest.json

@@ -0,0 +1,38 @@
+{
+  "README.md": {
+    "bytes": 3605,
+    "sha256": "3fbf8f22544665aa9f224094f65458dd71e50fc93500985a9a48f30f6d1df00b"
+  },
+  "atom_tokens.txt": {
+    "bytes": 28,
+    "sha256": "c7b8859e8b0fd79dfe48603073fcf10ded2d93da0e8de7a63fcc56515ab4d28f"
+  },
+  "charge_tokens.txt": {
+    "bytes": 7,
+    "sha256": "e8f533e310170a31a7e5e5fe125447fbe0253de205966ae295713c98c6ac531c"
+  },
+  "config.yaml": {
+    "bytes": 15569,
+    "sha256": "6cea95be3c2569771719e7313200e5c3c0e793a5c6cdb1db01898872b59263c7"
+  },
+  "distributions.pt": {
+    "bytes": 89925,
+    "sha256": "ba4101ee7f3284dd2087b15573f1730aaa70ee808d874ced396fcd64d55a66d5"
+  },
+  "edge_tokens.txt": {
+    "bytes": 16,
+    "sha256": "42b635ef938d2ec5a601044e4b01f3f84ed121f1bfaac984d86f180a8ccba3fd"
+  },
+  "epoch_100.ckpt": {
+    "bytes": 625018500,
+    "sha256": "4d17db9a77e75e12c452f0512301e5894b2d1a0bd9725186e35bf183a238b2b6"
+  },
+  "metadata.json": {
+    "bytes": 2175,
+    "sha256": "51b30352101276d7751d9d146e022c8e493d96695b7055cd11eb7c8f26f77ca5"
+  },
+  "summary.csv": {
+    "bytes": 3920,
+    "sha256": "7a1a9ce90367da79d59717dd158996992f7f0a2bc917449a1c3acb3582b3740b"
+  }
+}

crossdocked/scaffold/metadata.json

@@ -0,0 +1,89 @@
+{
+  "model": "crossdocked-scaffold",
+  "task": "scaffold_decoration_in_pocket",
+  "dataset": "CrossDocked",
+  "variant": "implicit_h",
+  "hydrogens": "implicit",
+  "completed_epochs": 100,
+  "global_step": 40000,
+  "checkpoint": "epoch_100.ckpt",
+  "checkpoint_sha256": "4d17db9a77e75e12c452f0512301e5894b2d1a0bd9725186e35bf183a238b2b6",
+  "checkpoint_bytes": 625018500,
+  "lora_rank": 8,
+  "lora_alpha": 8.0,
+  "base_model": "chembl3d-implicit-h",
+  "base_completed_epochs": 365,
+  "finetune_fraction": 0.9,
+  "in_vacuum_replay_fraction": 0.1,
+  "training_scaffold_kind": "libinvent",
+  "freeze_keep_positions": false,
+  "freeze_keep_charges": true,
+  "default_eval_weights": "ema",
+  "integration_steps": 100,
+  "original_checkpoint_including_optimizer": true,
+  "evaluation": {
+    "split": "test",
+    "protocol": "diffdec",
+    "panels": {
+      "single": 43,
+      "multi": 102
+    },
+    "seeds": [
+      42,
+      43,
+      44
+    ],
+    "samples_per_condition": 100,
+    "total_generation_attempts": 43500,
+    "precision": "float32 without autocast",
+    "standard_deviation": "sample (ddof=1)",
+    "summary": "summary.csv",
+    "compatibility_notes": [
+      "Published DiffDec conditions and metric formulas; original reference CSVs unavailable, so reference scaffold/protein strings are reconstructed.",
+      "Morph predicted bond topology is retained, rather than inferred from XYZ with Open Babel.",
+      "RDKit 2025.9.6 and deterministic per-attempt embedding seeds; upstream used older RDKit and unspecified embedding seeds.",
+      "Docking reconstructs and UFF-minimizes SMILES; the stored generated coordinates are unchanged."
+    ]
+  },
+  "atom_tokens": [
+    "B",
+    "Br",
+    "C",
+    "Cl",
+    "F",
+    "I",
+    "N",
+    "O",
+    "P",
+    "S",
+    "Se",
+    "Si"
+  ],
+  "edge_tokens": [
+    "<NO_BOND>",
+    "1",
+    "2",
+    "3"
+  ],
+  "charge_tokens": [
+    "-1",
+    "0",
+    "1"
+  ],
+  "pocket_vocab": {
+    "atom_tokens": [
+      "C",
+      "N",
+      "O",
+      "S"
+    ],
+    "edge_tokens": [
+      "<NO_BOND>"
+    ],
+    "charge_tokens": [
+      "0"
+    ]
+  },
+  "checkpoint_stored_epoch": 100,
+  "not_unconditional_catalog_entry": true
+}

crossdocked/scaffold/summary.csv

@@ -0,0 +1,11 @@
+model,epoch,panel,row_type,seed,weights,integration_steps,n_conditions,n_attempts,validity,uniqueness,similarity,recovery,chemistry_validity_exported_component_percent,chemistry_validity_exported_component_diffdec_weighted_percent,diffdec_validity_pooled_percent,diffdec_rejected_identical_scaffold_percent,diffdec_rejected_missing_scaffold_percent,diffdec_valid_chemistry_invalid_count,vina_score,reference_vina_score,high_affinity,docking_attempts,docking_successes,docking_failures,docking_timeouts,reference_docking_attempts,reference_docking_successes,invalid_predictions_replaced,high_affinity_denominator,docking_scored_groups
+crossdocked-scaffold,100,single,seed,42,ema,100,43,4300,88.29333333333334,52.160000000000004,0.8201344609798162,38.333333333333336,92.55813953488372,90.17333333333333,91.34883720930233,1.96,9.746666666666668,10,-8.033422353523228,-8.272,0.3995,4300,4282,18,0,43,43,372,4300,25
+crossdocked-scaffold,100,single,seed,43,ema,100,43,4300,89.53999999999999,54.076666666666675,0.8151260011360819,40.333333333333336,93.55813953488372,91.71333333333334,92.11627906976744,2.12,8.34,5,-8.04678868797334,-8.199333333333334,0.46490000000000004,4300,4286,14,0,43,43,339,4300,25
+crossdocked-scaffold,100,single,seed,44,ema,100,43,4300,88.31333333333333,53.72,0.8118672428330638,38.333333333333336,92.30232558139535,90.52333333333334,90.86046511627906,2.2800000000000002,9.406666666666668,8,-8.042993739683991,-8.128,0.4611333333333333,4300,4277,23,0,43,43,393,4300,25
+crossdocked-scaffold,100,multi,seed,42,ema,100,102,10200,89.94150782013686,66.77343597262953,0.7637482772393612,32.28005865102639,92.86274509803921,91.88801319648093,91.86274509803921,1.8451612903225807,8.213330889540567,76,-8.097944471948894,-8.16168970814132,0.43488863287250384,10200,10078,122,0,102,102,830,10200,31
+crossdocked-scaffold,100,multi,seed,43,ema,100,102,10200,89.63411534701858,66.49635874877809,0.7674004125271859,38.301564027370475,92.3529411764706,91.17148093841642,91.75490196078432,1.5580645161290323,8.807820136852396,87,-8.078235112509262,-8.61673579109063,0.3721236559139785,10200,10089,111,0,102,102,841,10200,31
+crossdocked-scaffold,100,multi,seed,44,ema,100,102,10200,89.83097507331378,66.02837243401758,0.7701394214114178,42.898338220918866,92.40196078431373,91.74369501466275,91.44117647058823,1.6419354838709674,8.52708944281525,63,-8.0883624708498,-8.491228878648235,0.4027611367127496,10200,10099,101,0,102,102,873,10200,31
+crossdocked-scaffold,100,single,mean,,ema,100,43,4300,88.71555555555555,53.31888888888889,0.8157092349829873,39.0,92.8062015503876,90.80333333333334,91.44186046511628,2.12,9.164444444444445,7.666666666666667,-8.04106826039352,-8.199777777777777,0.44184444444444443,4300,4281.666666666667,18.333333333333332,0,43,43,368,4300,25
+crossdocked-scaffold,100,single,std,,ema,100,43,4300,0.7140598584722949,1.0193480123933985,0.004164354122985984,1.1547005383792515,0.663640146537193,0.8072793816269578,0.633053841340294,0.16000000000000014,0.7339492363109881,2.516611478423583,0.00688805684003741,0.07200102879923423,0.03671969397919766,0.0,4.509249752822894,4.509249752822894,0.0,0.0,0.0,27.2213151776324,0.0,0.0
+crossdocked-scaffold,100,multi,mean,,ema,100,102,10200,89.80219941348975,66.43272238514173,0.7670960370593216,37.826653633105245,92.53921568627452,91.60106304985337,91.68627450980392,1.6817204301075268,8.516080156402737,75.33333333333333,-8.088180685102651,-8.423218125960062,0.40325780849974396,10200,10088.666666666666,111.33333333333333,0,102,102,848,10200,31
+crossdocked-scaffold,100,multi,std,,ema,100,102,10200,0.15570344274027922,0.3765861245782725,0.0032064254671021605,5.325046494448182,0.28125467230942774,0.37896243514468936,0.21900301866361663,0.14762544123005955,0.29739749424437123,12.013880860626733,0.009855937140807926,0.23502304411011748,0.031385436038058344,0.0,10.503967504392486,10.503967504392486,0.0,0.0,0.0,22.338307903688676,0.0,0.0

geom/scaffold_murcko_substituent_lora_r16/README.md

@@ -1,4 +1,6 @@
-# GEOM scaffold decoration: 25 completed fine-tuning epochs
+# geom-scaffold-murcko
+
+GEOM Murcko scaffold decoration after 25 completed fine-tuning epochs.
 
 This is the original checkpoint from the latest GEOM Murcko scaffold fine-tuning run,
 with rank-16 LoRA, trainable output heads, and substituent-constrained insertion

geom/scaffold_murcko_substituent_lora_r16/manifest.json

@@ -1,7 +1,7 @@
 {
   "README.md": {
-    "bytes": 2752,
-    "sha256": "3879125c6ede15b8b268f316c5c2163b1af0fb795582c3e4685acb029f52b74f"
+    "bytes": 2787,
+    "sha256": "887e268373f5e6e26bfbcbe0b429e4c6131c86ba18189d2a02007072fdb2ddd8"
   },
   "atom_tokens.txt": {
     "bytes": 22,
@@ -36,8 +36,8 @@
     "sha256": "ad57177b2289610ec10983c9f20480adca7e393166ffec78e9641d3546bc433c"
   },
   "metadata.json": {
-    "bytes": 1511,
-    "sha256": "f1200bd618873ce5ba3e1d0a3569cd852c6b1fbe6c7523863dfb567541228c1d"
+    "bytes": 1490,
+    "sha256": "8809046631ae6184775a7722ae64971fa9fa9406a3d0ec1b9733be690dca262a"
   },
   "panels/test_murcko_512_seed42.json": {
     "bytes": 226522,

geom/scaffold_murcko_substituent_lora_r16/metadata.json

@@ -9,7 +9,7 @@
   "checkpoint_bytes": 638268597,
   "original_checkpoint": "/cluster/project/krause/frankem/morph-chem/logs/training/geom-scaffold-substituent-lora-r16-20261012T095326Z/run/every_n_epochs/epoch=24-step=199100.ckpt",
   "training_job_id": 16886672,
-  "model": "geom-scaffold-murcko-substituent-lora-r16",
+  "model": "geom-scaffold-murcko",
   "lora_rank": 16,
   "lora_alpha": 16,
   "base_model": "geom-explicit-h",