genenmf-metaprogram-discovery

Cross-sample gene programs can be discovered by performing per-sample NMF to learn programs in each donor, then deriving consensus meta-programs. This per-sample factorization followed by consensus scoring enables cross-patient signatures without joint modeling.

A consensus meta-program is a reproducible transcriptional program identified by performing per-sample non-negative matrix factorization across multiple donors, then deriving shared programs through consensus scoring. It captures cross-sample gene signatures without requiring a joint model.

Split your Seurat object by donor, run multiNMF across samples to learn gene programs in each, then extract and derive consensus meta-programs. This workflow uses GeneNMF to enable cross-sample meta-program discovery.

Yes, per-sample NMF is specifically suited for multi-donor tumor datasets where batch effects hinder joint modeling. It factorizes each sample independently, then uses consensus scoring to find reproducible meta-programs across donors without batch correction.

After extracting consensus meta-programs from per-sample NMF, use UCell within Seurat workflows to score cells by meta-program signatures. This enables quantification of program activity across donors in tumor heterogeneity studies.

Per-sample NMF is preferred when batch effects across patients or donors are too strong for joint modeling. It avoids batch correction by factorizing each sample independently and deriving consensus meta-programs through cross-sample scoring.