Abstract

Accurately predicting chromatin loops from genome-wide interaction matrices such as Hi-C data is critical to deepen our understanding of proper gene regulation events. Current approaches are mainly focused on searching for statistically enriched dots on a genome-wide map. However, given the availability of a wide variety of orthogonal data types such as ChIA-PET, GAM, SPRITE, and high-throughput imaging, a supervised learning approach could facilitate the discovery of a comprehensive set of chromatin interactions. Here we present Peakachu, a Random Forest classification framework that predicts chromatin loops from genome-wide contact maps. Compared with current enrichment-based approaches, Peakachu identified more meaningful short-range interactions. We show that our models perform well in different platforms such as Hi-C, Micro-C, and DNA SPRITE, across different sequencing depths, and across different species. We applied this framework to systematically predict chromatin loops in 56 Hi-C datasets, and the results are available at the 3D Genome Browser (www.3dgenome.org).