Rxivist logo

Integrating distal and proximal information to predict gene expression via a densely con-nected convolutional neural network

By Wanwen Zeng, Yong Wang, Rui Jiang

Posted 08 Jun 2018
bioRxiv DOI: 10.1101/341214 (published DOI: 10.1093/bioinformatics/btz562)

Motivation: Interactions among such cis-regulatory elements as enhancers and promoters are main driving forces shaping context-specific chromatin structure and gene expression. Although there have been computational methods for predicting gene expression from genomic and epige-nomic information, most of them overlook long-range enhancer-promoter interactions, due to the difficulty in precisely linking regulatory enhancers to target genes. Recently, a novel high-throughput experimental approach named HiChIP has been developed and generating compre-hensive data on high-resolution interactions between promoters and distal enhancers. On the other hand, plenty of studies have suggested that deep learning achieves state-of-the-art perfor-mance in epigenomic signal prediction, and thus promoting the understanding of regulatory ele-ments. In consideration of these two factors, we integrate proximal promoter sequences and HiChIP distal enhancer-promoter interactions to accurately model gene expression. Results: We propose DeepExpression, a densely connected convolutional neural network to pre-dict gene expression using both promoter sequences and enhancer-promoter interactions. We demonstrate that our model consistently outperforms baseline methods not only in the classifica-tion of binary gene expression status but also in the regression of continuous gene expression levels, in both cross-validation experiments and cross-cell lines predictions. We show that se-quential promoter information is more informative than experimental enhancer information while enhancer-promoter interactions are most beneficial from those within +-1000 kbp around the TSS of a gene. We finally visualize motifs in both promoter and enhancer regions and show the match of identified sequence signatures and known motifs. We expect to see a wide spectrum of appli-cations using HiChIP data in deciphering the mechanism of gene regulation.

Download data

  • Downloaded 684 times
  • Download rankings, all-time:
    • Site-wide: 45,057
    • In bioinformatics: 4,717
  • Year to date:
    • Site-wide: 67,372
  • Since beginning of last month:
    • Site-wide: 52,272

Altmetric data

Downloads over time

Distribution of downloads per paper, site-wide