Add the cNF organoid dataset: samples and omics

[jjacobson95]

Pipeline Hardening/Debugging PR # 7

Add the cNF organoid dataset: samples and omics

Adds the first half of a new dataset, cNF (cutaneous neurofibroma), a patient-derived-organoid drug screen with matched multi-omics. This PR covers specimen/sample generation and all three omics types plus a shared helper module and container image. The drug table and dose-response experiments follow in the next PR.

Samples (01-samples-cnf.py)

• Gathers specimens from four Synapse sources: the drug-screen file index (syn51301431), the global proteomics matrix (syn74815895), the RNA discovery file (syn71333780), and the Normal Skin folder (syn74284682). Each source Synapse ID is overridable via environment variable.
• Classifies each specimen and assigns the matching schema model type:
  • organoid and drug-treated organoid map to "patient derived organoid" (treated IDs keep their treatment label, for example NF0021_T1_Onalespid_1uM);
  • tumor tissue maps to "tumor";
  • normal skin maps to "normal_tissue".
• Keeps tumor tissue and organoids from the same tumor as separate rows with distinct sample IDs, because they are different biology and different model types; treated and untreated organoids are likewise separate rows.
• Continues improve_sample_id from the previous samples file (max + 1) so IDs don't collide across datasets.
• Deliberately omits a per-sample cohort label, because the upstream omics is batch-corrected and cohort membership disagrees across modalities; a uniform project label is stored instead.

Omics (02-omics-cnf.py)

• Transcriptomics: gene-level TPM assembled from per-cohort salmon gene-TPM matrices (default cohorts syn66352931 and syn70765053; overridable). Protocol-optimization samples are excluded.
• Proteomics: batch-corrected global proteomics from syn74815895 (the correctedAbundance values).
• Phosphoproteomics: batch-corrected phosphoproteomics from syn70078415, mapped to phosphosite_id values using the phosphosites.csv reference from the previous PR.
• All specimen names are canonicalized through the shared helper so values like NF0018.T1.organoid resolve to the canonical IDs in cnf_samples.csv; source rows that don't match a known sample are dropped via an inner join.

Shared helper and packaging

• cnf_utils.py: specimen classification and canonicalization (classify_specimen, patient_from_specimen, canonicalize_specimen_column) shared across the samples, omics, and (next PR) experiments steps.
• Build entrypoints build_samples.sh and build_omics.sh, requirements.txt, the container image Dockerfile.cnf, and a README.md documenting sources and the specimen model.

Dependencies and build policy

• Depends on the phosphosites PR (phospho mapping needs phosphosites.csv).
• cNF is a fully buildable dataset (--dataset cnf) but is intentionally not part of the default build list (set in the later build-orchestration PR).

Scope: 8 files, all new. Base: phosphosites-reference-builder.

[sgosline]

I thought we weren't including the cNF data?

[jjacobson95]

These are currently locked behind Synapse permissions. They are located in the "Leveraging patient-derived cutaneous neurofibroma organoid models to identify biomarkers of drug response" Project (syn51301431).

Keep them locked until we want to release.