PRISM-GEP: Viewing Single-Cell Expression through the Lens of Topic Modeling.

PRISM-GEP uncovers overlapping gene-expression programs in single-cell RNA-seq data with performance comparable to specialized methods across 15 datasets.

• Why it matters: Understanding co-regulated gene modules is crucial for decoding biological processes, but existing methods often fail to capture overlapping programs due to their one-to-one gene assignments.
• What they did: The approach applies Latent Dirichlet Allocation with a gene co-expression prior, analyzing 15 human and mouse datasets without requiring reference datasets or pathway databases.
• The result: PRISM-GEP matches top-performing methods on Gene Ontology metrics, consistently ranks highly, and successfully recovers developmental gene cascades, enabling more accurate biological insights.