Physics of the living genome

The long genome — roughly one meter in a human cell — undergoes severe compaction to fit inside a nucleus only ten microns across. Rather than collapsing into a featureless globule, the genome self-organizes into a rich hierarchical structure: chromosomes occupy separate territories, compartments of open and closed chromatin are interspersed, and topological domains partition the chromosome at the megabase scale.

Our approach is grounded in the polymeric nature of the genome. We microscopically model living DNA as a twistable semiflexible polymer where protein-mediated kinks and bends act as sinks or sources of bending and twisting energies at hundreds of base pairs. Our coarse-grained view, considering ~50 kb units, is that of a partially collapsed polymer where active mechanisms drive 3D contacts between far-away genomic units. We have shown that effective representations can capture many aspects of genome organization. Our goal is to learn the physics of these active mechanisms and decipher the regulatory code of the living genome.

Physics-based AI

Pure data-driven AI models lack physical interpretability; pure physics models struggle to capture the complexity of biological systems. We are building AI models that are grounded in physical principles — incorporating known symmetries, conservation laws, and mechanistic constraints — to predict genome structure, infer model parameters from experimental data, and generate physically meaningful insights.