Detecting Community Structures in Hi-C Genomic Data

TL;DR

This paper introduces a community detection approach for Hi-C genomic data that identifies variable and multi-scale DNA communities, revealing both known and novel structural features of the genome.

Contribution

It presents a flexible community detection method that can determine the number of communities and detect both large and small-scale DNA structures in Hi-C data.

Findings

Successfully recovers known DNA communities with k=2

Identifies an optimal k=6 revealing new communities

Detects topologically associating domains (TADs) at multiple scales

Abstract

Community detection (CD) algorithms are applied to Hi-C data to discover new communities of loci in the 3D conformation of human and mouse DNA. We find that CD has some distinct advantages over pre-existing methods: (1) it is capable of finding a variable number of communities, (2) it can detect communities of DNA loci either adjacent or distant in the 1D sequence, and (3) it allows us to obtain a principled value of k, the number of communities present. Forcing k = 2, our method recovers earlier findings of Lieberman-Aiden, et al. (2009), but letting k be a parameter, our method obtains as optimal value k = 6, discovering new candidate communities. In addition to discovering large communities that partition entire chromosomes, we also show that CD can detect small-scale topologically associating domains (TADs) such as those found in Dixon, et al. (2012). CD thus provides a natural and flexible statistical framework for understanding the folding structure of DNA at multiple scales in Hi-C data.