Generation of Hi-C Maps from DNA Sequence Data Using Deep Learning

Information about chromosome conformation is important for understanding the regulation of gene expression, DNA replication, reparation and recombination. Hi-C is an experimental technique that allows one to assay chromatin DNA interactions with high resolution and coverage. On the other hand, in silico methods are emerging to predict chromosome conformation from DNA sequence alone. Currently, these methods exist as both proofs of concept and, most importantly, as tools for advanced data interpretation and feature importance analysis. However, most of these methods have been developed specifically for the mammalian chromatin (human and mouse). Here we report a Convolutional Neural Network architecture to reconstruct a Hi-C map data given a DNA sequence of insect Drosophila melanogaster. We discuss possible improvements that might lead to robust Hi-C reconstruction from genomic data and biologically relevant interpretation of DNA features.