Machine learning of enhancer-promoter specificity based on enhancer perturbation studies reveals a distinct class of enhancers.

Abstract

MotivationUnderstanding the rules that govern enhancer-driven transcription remains a central unsolved problem in genomics. Now with multiple massively parallel enhancer perturbation assays published, there are enough data that we can utilize to learn to predict enhancer promoter relationships in a data driven manner. ResultsWe applied machine learning to one of the largest enhancer perturbation studies integrated with transcription factor and histone modification ChIP-seq. Based on the learned model, we confirmed previously reported rules governing enhancer driven transcription, and we gained some insights that generated new hypotheses, such as a novel role for protecting against replication-transcription conflict at the active enhancers in CHAMP1. We also identified a distinct class of enhancers that drives target promoter transcription, but is not in strong contact with the promoters. There were two clusters of such enhancers that regulated ATG2A and the histone 1 cluster respectively. These enhancers were different from other typical enhancers, in that they had other strong enhancers nearby, and they also had strong H3K4me3 marks at the target promoters, both patterns that typically predict reduced enhancer influence, but here contributing in the opposite way. In summary, we find that integrating genomic assays with enhancer perturbation studies increases the accuracy of the model, and provides novel insights into the understanding of enhancer driven transcription. Availabilitythe trained models and the source code are available at https://github.com/HanLabUNLV/abic. Contactmira.han@unlv.edu