VIGAT: VLM-guided gene activity quantification from single-cell ATAC-seq

VIGAT, a VLM-guided tool for single-cell ATAC-seq, outperforms five existing methods across 196 paired RNA-seq/ATAC-seq datasets in recovering marker-associated accessibility enriched in matching cell populations.

• Why it matters: Gene activity inference from single-cell ATAC-seq is hard because informative regulatory regions differ by gene and tissue, and using broadly accessible regions can yield weak cell-type discrimination.
• What they did: VIGAT uses a vision-language model to interpret ATAC coverage plots, selecting gene-specific genomic regions enriched for cluster-discriminative signals by integrating annotations and biological context, then builds a gene-by-cell activity matrix.
• The result: In 196 paired single-cell RNA-seq and ATAC-seq datasets, VIGAT recovers marker-associated accessibility enriched in corresponding cell populations better than five prior methods, supporting scalable, visual-reasoning-based gene activity quantification.

AI summary of the abstract · may contain errors · not medical advice · read the paper.