Chin, W. L. | bioRxiv Science

bioRxiv2025

GeneInsight: Condensing Gene Set Knowledge via Language Models

Interpreting gene sets is often complicated by the overwhelming number of annotations associated with individual genes, making it difficult to extract meaningful biological insights. To address this issue, we developed GeneInsight, an AI-powered tool that combines advanced topic modelling with large language models to automatically synthesise diverse biological annotations from literature, gene ontologies, and databases such as STRING. GeneInsight consolidates extensive annotations into coherent thematic summaries that render such data readily interpretable, thereby enabling the rapid extraction of biologically significant insights that conventional enrichment analyses often overlook.

bioinformatics↗

bioRxiv2023

Spectral detection of condition-specific biological pathways in single-cell gene expression data

Single cell RNA sequencing is an ubiquitous method for studying changes in cellular states within and across conditions. Differential expression (DE) analysis may miss subtle differences, especially where transcriptional variability is not unique to a specific condition, but shared across multiple conditions or phenotypes. Here, we present CDR-g (Concatenate-Decompose-Rotate genomics), a fast and scalable strategy based on spectral factorisation of gene coexpression matrices. CDR-g detects subtle changes in gene coexpression across a continuum of biological states in multi-condition single cell data. CDR-g collates these changes and builds a detailed profile of differential cell states. Applying CDR-g, we show that it identifies biological pathways not detected using conventional DE analysis and delineates novel, condition-specific subpopulations in single-cell datasets.

bioinformatics↗