CATaN maps gene regulatory programs that shape genetic risk across complex diseases

Abstract

Causal variants of complex traits are enriched at transcription factor (TF) binding sites and are thought to contribute to pathology by disrupting TF activity and thereby causing transcriptome dysregulation. However, existing approaches typically address TF-mediated gene regulatory networks (TF-GRNs) and transcriptomes separately, and methods that jointly leverage both to systematically assess disease heritability remain limited. We aimed to develop a framework that jointly leverages TF-GRNs and transcriptomes to assess disease heritability. Here, we constructed a matrix encoding TF-GRNs and developed an unsupervised analytical pipeline, Canonical correlation Analysis of Transcriptome and TF-gene regulatory Networks (CATaN). CATaN applies canonical correlation analysis (CCA) to extract canonical correlation (CC) components, i.e., shared variation components between transcriptomes and TF-GRNs, and converts them into genome-wide functional annotation scores connected to stratified LD score regression (S-LDSC) for heritability analysis. We applied CATaN to eight datasets, including 19,198 bulk samples and 611,772 single cells from human and mouse sources, identifying 588 CC components that are significantly enriched for SNP heritability across 69 complex traits. Notably, functional annotation tracks based on these TF-GRNs are distinct from transcriptome signatures prioritized by LDSC-SEG, with greater heritability enrichment for a subset of traits. Finally, we suggest that CATaN may help prioritize candidate causal variants for experimental fine-mapping using genome editing. Together, integrating TF-GRNs with transcriptomes reveals disease-relevant regulatory programs that are not fully captured by transcriptome-based analyses alone.