py3ca

Python reimplementation of the 3CA NMF pipeline (Gavish et al., 2023).

What it does

• Per-sample NMF (K=4..9) after 3CA preprocessing.
• Robust program filtering (within-sample, across-sample, non-redundant).
• Greedy clustering into Meta-Programs (MPs).
• MP scoring per cell with one-sample t-test and FDR correction.
• Program scoring per cell from NMF loadings (per sample).

Quick start

1. Install dependencies:

• See requirements.txt

2. Run:

• python run.py --data-dir /mnt/sdb/xzh/Vproject/TCA/data --out-dir /mnt/sdb/xzh/Vproject/TCA/py3ca/out

Outputs are written under the output directory.

Output highlights

• scores/<sample_id>_mp_scores.csv: MP-level cell scores (t_stat, p_value, fdr).
• program_scores/<sample_id>_program_scores.csv: program-level cell scores from NMF W loadings.