Cheap Map: Hi-C from ChIP-Seq through Machine Learning

In the nucleus of eukaryotic cells, chromatin is organized in specific conformations which depend on cell type; these conformations have been most recently studied through DNA-DNA ligation assays. Obtaining these data about the three-dimensional structure of the chromosome is an expensive and time consuming process. We exploit the idea that epigenetic data determine chromatin architecture by developing a tool that can predict the chromatin contact map purely from epigenetic data without using any structural information. The model has been trained on chromosome 1 and 2 and can quickly predict the high-resolution contact maps (Hi-C) of chromosome 1-22 for the human lymphoblastoid cell line. This project was completed with contributions from Jose Onuchic from the Rice University Chemistry department.