Decorrelating Hi-C contacts enhances identification of the interactions driving genomic structure and organization.

Abstract

Three-dimensional genomic organization underpins gene expression and cell differentiation, essential processes for multicellular life with potential applications to human medicine. This has motivated the development of many experimental and theoretical approaches to identify structural features, understand folding mechanisms, and predict genomic organization from sequencing data. In particular, high-throughput chromosome conformation capture experiments (Hi-C) provide detailed in vivo structural information. Two disparate processes contribute to the observed Hi-C contact densities: Some contacts are energetically favorable due to biologically important interactions; other contacts are entropically favorable due to the polymer structure of chromosomes. These effects overlap in Hi-C data, so differentiating between structure-driving interactions and polymer-driven correlations as the primary cause of individual high-probability contacts remains challenging. To address this, we developed a rigorous statistical mechanical model that separately accounts for interaction energies and polymer topology. We then efficiently invert this model to convert correlated Hi-C contact probabilities into uncorrelated contact energies, revealing the genomic location of structurally important interactions. Comparing these locations to one-dimensional genetic and epigenetic data will reveal unknown genomic folding mechanisms. This mechanistic insight will improve the accuracy of contact energy predictions from one-dimensional sequence information. Furthermore, our model can convert contact energy predictions into contact probabilities, providing a novel approach to de novo genomic organization predictions.