Accurate prediction of cis-regulatory modules reveals a prevalent regulatory genome of humans

Abstract

Annotating all cis-regulatory modules (CRMs) and transcription factor (TF) binding sites(TFBSs) in genomes remains challenging. We tackled the task by integrating putative TFBSs motifs found in available 6,092 datasets covering 77.47% of the human genome. This approach enabled us to partition the covered genome regions into a CRM candidate (CRMC) set (56.84%) and a non-CRMC set (43.16%). Intriguingly, like known enhancers, the predicted 1,404,973 CRMCs are under strong evolutionary constraints, suggesting that they might be cis-regulator. In contrast, the non-CRMCs are largely selectively neutral, suggesting that they might not be cis-regulatory. Our method substantially outperforms three state-of-the-art methods (GeneHancers, EnhancerAtlas and ENCODE phase 3) for recalling VISTA enhancers and ClinVar variants, as well as by measurements of evolutionary constraints. We estimated that the human genome might encode about 1.46 million CRMs and 67 million TFBSs, comprising about 55% and 22% of the genome, respectively; for both of which, we predicted 80%. Therefore, the cis-regulatory genome appears to be more prevalent than originally thought.