GFETM: Genome foundation-based embedded topic model for scATAC-seq modeling (opens in new tab)

We develop GFETM, a framework integrating pre-trained genome foundation models with interpretable topic modeling to analyze single-cell chromatin accessibility data. GFETM achieves accurate cell clustering; enables transfer learning across tissues, species, and modalities; infers cell-type-specific transcription factor activities; and identifies disease-specific epigenomic signatures.