Title: Learning Principles of Genome Structure and Function from Minimalist 3D Polymer Models: Self-Avoidance, Volume Confinement, and Biological Interactions.
Speaker: Jie Liang, University of Illinois at Chicago
Hosted by: Pratyush Tiwary
Abstract:
3D organization of chromosomes in cell nucleus play important roles in genome functions. Essential information on chromatin structure has been gathered, largely from chromosome conformation capture (Hi-C), the experimental technique that measures pairwise interactions between genomic regions. Here we demonstrate how a minimalist 3D polymer model of chromatin can shed light on principles of genome organization. Our minimalist models are based on basic properties that 3D chromatin conformations: 1) must be self-avoiding, and satisfy 2) the constraint of cell nuclei volume and 3) the constraints from a small set of biological interactions (<5-13%). To obtain correctly packed 3D chromatin polymer conformations in confined nucleus volume, we have overcome technical challenges and developed deep sampling methods. These methods allow us to generate properly-sampled large ensembles of 3D chromatin conformations. Our minimalist models can explain: a) the physical basis of scaling relationship such as contact probability of two regions and their genomic separation, b) the origin of the structures of topologically-associating domains. Without invoking any adjustable parameter, our model can also c) fold chromatin chains into 3D conformations using the small set of biological interaction, achieving high accuracy (R=0.96-0.97). The single-cell 3D chromatin conformations in our model ensemble d) are in excellent agreement with single-cell experimental measurements, and e) can quantitatively characterize chromatin heterogeneity. We further discuss novel biological findings relating genome structures and genome function. These include orchestrated changes in chromatin heterogeneity during Drosophila embryogenesis, prominence of functional hubs of higher-order many-body interactions, their global landscape in active genomic regions, as well as the 1D signature of epigenetic modifications of many-body interactions as identified by machine learning. (See https://rdcu.be/cdgFv for a recent publication).