scROMA reproducibility workflow: analysis and figure-generation code

Analysis and figure-generation code reproducing every figure and table in the scROMA manuscript (Zhubanchaliyev et al.). scROMA is an SVD-based method for pathway-activity inference in single-cell transcriptomics providing per-cell activity scores, per-gene contributions, permutation-based significance, and a batch-aware extension that corrects batch effects within the gene-set subspace rather than across the full transcriptome. This record archives the workflow, not the tool. The scROMA package itself (v0.3.2, the version described in the manuscript) and the supporting data are archived separately; see Related works. Figures 3, 5 and 6 and all supplementary figures rebuild from a bare clone. Figures 4 and 7 additionally require one large derived object each: run python scripts/fetch_zenodo_data.py to retrieve them from the derived-data record. Environment: conda env create -f environment.yml, which installs scroma==0.3.2 from PyPI. anndata >= 0.12 is required — older versions cause the lung CNV figure to fall back silently to synthetic preview data. Seeds: simulation replicates 42-51; PBMC batch realizations 0-5; all other analyses 42. The one exception is the scANVI integration of the cystic fibrosis atlas, which sets no seed and is therefore deposited as a fixed object rather than regenerated.