Hybrid Machine Learning and Polymer Physics Approach to Investigate 3D Chromatin Structure

Innovative experimental protocols from Molecular Biology provided in recent years quantitative data about the structure of the cell nucleus. These technologies, such as Hi-C, GAM or SPRITE, revealed that the genome has a non-random three-dimensional (3D) spatial organization, which serves functional purposes. In order to dissect the complexity of chromosome folding, models from Polymer Physics have been employed, highlighting many key aspects of large-scale chromatin organization. A deep understanding of the molecular mechanisms underlying the genome architecture is currently a crucial problem in Biology, since chromatin misfolding or structural variants can reconfigure chromatin domains, thereby resulting in pathogenic phenotypes and disease. Here, we discuss a numerical Polymer-Physics-based approach (PRISMR), able to model 3D chromatin folding by using Machine Learning strategies informed with experimental data. Using as a case study the Pitx1 locus, a genomic region critically involved in hindlimb development, we show that the PRISMR algorithm reproduces in silico with high accuracy the experimental contact data, thus providing a powerful computational tool for analyzing and predicting the 3D chromatin structure.