Mapping the spectrum of 3D communities in human chromosome conformation capture data

TL;DR

This paper introduces a novel network community detection method that maps the continuous spectrum of 3D chromosomal structures in Hi-C data, revealing new insights into the organization of TADs and compartments.

Contribution

It presents a unified approach to detect multi-scale 3D chromosomal communities using a single resolution parameter, bridging TADs and compartments.

Findings

Communities can include non-consecutive polymer segments.

CTCF protein correlates with community borders at specific scales.

TADs and compartments are part of a continuous spectrum of 3D communities.

Abstract

Several experiments show that the three dimensional (3D) organization of chromosomes affects genetic processes such as transcription and gene regulation. To better understand this connection, researchers developed the Hi-C method that is able to detect the pairwise physical contacts of all chromosomal loci. The Hi-C data show that chromosomes are composed of 3D compartments that range over a variety of scales. However, it is challenging to systematically detect these cross-scale structures. Most studies have therefore designed methods for specific scales to study foremost topologically associated domains (TADs) and A/B compartments. To go beyond this limitation, we tailor a network community detection method that finds communities in compact fractal globule polymer systems. Our method allows us to continuously scan through all scales with a single resolution parameter. We found: (i) polymer segments belonging to the same 3D community do not have to be in consecutive order along the polymer chain. In other words, several TADs may belong to the same 3D community. (ii) CTCF proteins---a loop-stabilizing protein that is ascribed a big role in TAD formation---are well correlated with community borders only at one level of organization. (iii) TADs and A/B compartments are traditionally treated as two weakly related 3D structures and detected with different algorithms. With our method, we detect both by simply adjusting the resolution parameter. We therefore argue that they represent two specific levels of a continuous spectrum 3D communities, rather than seeing them as different structural entities.