|
Download docs/upstream.md from xuyun/CIP2A: direct link, hf CLI and curl.
- Browser
- Download file 3.69 kB
-
https://huggingface.co/datasets/xuyun/CIP2A/resolve/main/docs/upstream.md
- Command line
-
hf download hf://datasets/xuyun/CIP2A/docs/upstream.md
-
curl -L -o upstream.md https://huggingface.co/datasets/xuyun/CIP2A/resolve/main/docs/upstream.md
3.69 kB
| # Reanalysis from public inputs | |
| Start at the repository root. The CSVs in [provenance](../provenance/) give source URLs, the expected local filenames, and the fingerprints of the study inputs where established. Save them under `inputs/`. The DepMap release is identified by the DOI in the dataset card. Its manifest entry records the mirror copy actually used. The Cellosaurus results and UCSC tables may change over time. A checksum difference means the bytes differ from the recorded input. For GSE281740, the manifest points to GEO and captures the original archive fingerprint. | |
| ```text | |
| python -m pip install -r requirements.txt | |
| python -c "from pathlib import Path; [Path(p).mkdir(parents=True, exist_ok=True) for p in ('inputs','build/bulk','build/screens','build/chan','build/external')]" | |
| ``` | |
| ## Bulk tissue | |
| Obtain the four GEO files and two MSigDB gene-set files listed in `source_manifest.csv`. The gene-set files are required to reproduce the full six-feature adjustment families used in the original tissue analyses. | |
| ```text | |
| python scripts/check_inputs.py gse149507.csv.gz gse149507.meta.gz gse60052.tsv.gz gse60052.meta.gz hallmark.gmt macro.gmt | |
| python scripts/analyze.py inputs build/bulk --only tissue | |
| ``` | |
| Compare `build/bulk/tissue_effects.csv` and `build/bulk/tissue_sample_values.csv` with the corresponding bundled files. | |
| ## Genetic screens | |
| Obtain `CRISPRGeneEffect.csv`, all three recorded Cellosaurus query tables, and the Bioconductor ExperimentHub resource `rnai_19Q3.rda`. The latter is approximately 73 MB. The streaming reader extracts the relevant RNAi gene rows before the model join. | |
| ```text | |
| python scripts/check_inputs.py CRISPRGeneEffect.csv cello-exact-sclc.tsv cello-exact-luad.tsv cello-exact-luad-page2.tsv rnai_19Q3.rda | |
| python scripts/analyze.py inputs build/screens --only dependency --effects inputs/CRISPRGeneEffect.csv --model-source inputs --cellosaurus | |
| python scripts/stream_rnai.py inputs/rnai_19Q3.rda build/screens/rnai_targets.csv | |
| python scripts/demeter2.py build/screens/rnai_targets.csv build/screens/depmap_model_scores.csv build/screens | |
| ``` | |
| Compare the CRISPR outputs with `data/depmap_model_scores.csv` and `data/depmap_comparison.csv`. Filter `build/screens/demeter2_SCLC.csv` to the CIP2A gene for comparison with `data/rnai.csv`. | |
| ## Single-cell cases | |
| Obtain the Chan CELLxGENE H5AD release (~1.46 GB), the merged GSE281740 Cell Ranger file (~208 MB), and both GRCh38 annotation tables from the corresponding provenance entries. | |
| ```text | |
| python scripts/check_inputs.py chan.h5ad GSE281740_sclc_cellranger.csv.gz ncbiRefSeqCurated.txt.gz cytoBand.txt.gz | |
| python scripts/chan.py inputs/chan.h5ad build/chan --strict-only | |
| python scripts/extract_chan_umap.py inputs/chan.h5ad build/chan --strict-only | |
| python scripts/gse281740_extract.py inputs/GSE281740_sclc_cellranger.csv.gz build/external | |
| python scripts/cnv_arm_audit.py build/external inputs/ncbiRefSeqCurated.txt.gz inputs/cytoBand.txt.gz build/external | |
| python scripts/cip2a_stability.py build/external inputs/ncbiRefSeqCurated.txt.gz inputs/cytoBand.txt.gz build/external | |
| ``` | |
| The Chan summaries use published annotations for the four confirmed primary lung cases. The GSE281740 extraction uses RU1449A (GSM8627394) and RU1334 (GSM8627403). The other primary lung specimens in GSE281740, RU1108A, RU426B and RU1195A, share patient identifiers with the Chan atlas. The external classifier uses both sample-specific reference groups, with counts and CPM summarized after assignment. Compare the aggregate files and stability grid with their counterparts in `data/`. Upstream reanalysis writes working intermediates such as barcode-level assignments under `build/` for local use. | |