HilbertEPIs: Enhancer-Promoter Interactions Prediction with Hilbert Curve and CNN Model

Abstract

Enhancers are DNA cis-regulatory sequences that control the transcriptional activities of many gene regulation elements. Due to enhancers always get close to promoters by complex spatial structures, accurately identifying Enhancer-Promoter Interactions will help us understand mechanisms of gene regulations, recognize specific genes associated with diseases, as well as offer help with disease diagnosis and treatment. In this article, we develop a model named HilbertEPIs to predict the interactions between enhancers and promoters. We first transfer 1D sequence into 3D picture representations with Hilbert Curve to preserve the spatial structure of this sequence. Then extract features by CNN model. Finally, using two strategies to deal with unbalanced data. Experimental results have proved that HilbertEPIs has perfect performance compared to existed methods, as well as to show that Hilbert Curve is qualified to represent spatial relationships among different genetic regulatory elements. We train model in two ways and learn from six cell lines, finally achieve the data in 0.908~0.983 of AUROC, 0.926~0.988 of AUPR.