Abstract:Enhancer-promoter interaction (EPI) is an important cis-regulatory mechanism in the regulation of tissue-specific gene expression. However, it still has limitation to precisely identity these interactions so far. In this paper, using diverse genomic features for various regulatory regions, we presented a computational approach to predict EPIs with improved accuracies. Meanwhile, we comprehensively studied more potential regulatory factors that are important to EPIs prediction, such as nucleosome occupancy, enhancer RNA; and found the cell line-specificity and region-specificity of the contributions of diverse regulatory signatures. By adding genomic signatures of segmented regulatory regions, our best accuracies of cross-validation test were about 11%–16% higher than the previous results, indicating the location-specificity of genomic signatures in a regulatory region for predicting EPIs. Additionally, more training samples and related features can provide reliable performances in new cell lines. Consequently, our study provided additional insights into the roles of diverse signature features for predicting long-range EPIs.

Recognition of long-range enhancer-promoter interactions by adding genomic signatures of segmented regulatory regions.