Li Chen, Ph.D.
Associate Professor
Department of Biostatistics
University of Florida
Abstract. Understanding gene regulatory networks (GRNs) is essential for deciphering biological processes and disease mechanisms. Single-cell multiome technologies enable joint profiling of chromatin accessibility and gene expression, providing a powerful framework for inferring cell type–specific GRNs. However, existing methods typically analyze cell types independently or rely on pseudo-bulk aggregation, limiting their ability to resolve rare populations and capture cellular heterogeneity. Here, we introduce BayesCNet, a Bayesian hierarchical model that jointly infers enhancer-gene linkages across all cell types by leveraging cell type relationships, either curated from biological knowledge or inferred from data-driven cell representation, to enable information sharing. On synthetic data, BayesCNet consistently outperforms state-of-the-art methods, with the largest gains in rare populations. Applied to PBMC and hematopoietic single-cell datasets, BayesCNet yields EG linkages with higher concordant to cell type–specific promoter-capture Hi-C interactions. Stratified LD score regression further demonstrated that BayesCNet-identified enhancers exhibited stronger enrichment for trait heritability, while colocalization analyses showed that BayesCNet preferentially captures disease-associated GWAS SNPs within EG linkages at key immune and hematopoietic genes. Finally, BayesCNet reconstructs cell type–specific GRNs that reveal shared, cell type–specific and cooperative regulators, recovering regulatory programs consistent with known biological functions.