Simulating topological domains in human chromosomes with a fitting-free model

TL;DR

This paper introduces a simple, fitting-free polymer model that simulates the 3D organization of human chromosomes, accurately predicting topological domains and boundaries without relying on Hi-C data.

Contribution

The model is novel in its ability to predict chromosome organization without using experimental Hi-C data as input.

Findings

Successfully predicts most TAD boundaries in simulations

Reproduces key features of chromosome folding

Operates without fitted parameters from Hi-C data

Abstract

We discuss a polymer model for the 3D organization of human chromosomes. A chromosome is represented by a string of beads, with each bead being "colored" according to 1D bioinformatic data (e.g., chromatin state, histone modification, GC content). Individual spheres (representing bi- and multi-valent transcription factors) can bind reversibly and selectively to beads with the appropriate color. During molecular dynamics simulations, the factors bind, and the string spontaneously folds into loops, rosettes, and topologically-associating domains (TADs). This organization occurs in the absence of any specified interactions between distant DNA segments, or between transcription factors. A comparison with Hi-C data shows that simulations predict the location of most boundaries between TADs correctly. The model is "fitting-free" in the sense that it does not use Hi-C data as an input; consequently, one of its strengths is that it can -- in principle -- be used to predict the 3D organization of any region of interest, or whole chromosome, in a given organism, or cell line, in the absence of existing Hi-C data. We discuss how this simple model might be refined to include more transcription factors and binding sites, and to correctly predict contacts between convergent CTCF binding sites.