TFXcan reveals transcription factor programs driving GWAS loci

Disease-associated loci identified by genome-wide association studies (GWASs) may induce effects by altering transcription factor (TF) binding. However, due to the scarcity of population-level TF-binding experiments, identifying these disrupted TFs remains challenging. To address this, we present the TFXcan framework, which correlates genetically predicted TF binding with phenotypes to nominate TF-tissue pairs that may regulate complex traits. Within this framework, we introduce Enpact, a deep learning-based approach that integrates Enformer and IMPACT to infer personalized TF-binding events. Leveraging Enpact, we correlate predicted binding profiles for 700 TF-tissue pairs with prostate cancer risk, identifying multiple implicated TFs at risk loci. Matrix decomposition further reveals clusters of TF-tissue pairs, termed "programs," acting as candidate drivers of disease risk. TFXcan represents the first method capable of predicting TF binding from genotype data at scale, enabling systematic identification of TFs and regulatory programs underlying complex traits.