Physics-Based AI for Modeling 3D Genome Organization

Three-dimensional genome organization plays a central role in gene regulation and

cell identity, yet predicting chromatin folding across biological contexts remains a

major challenge. In this talk, we will discuss a physics-based AI framework for

modeling chromosomal ensembles by combining polymer physics, maximum-entropy

inference, and modern machine learning architectures. Using data-driven energy

functions inferred from Hi-C and high-resolution DNA-tracing microscopy, this

framework enables chromatin dynamics simulations that reproduce experimental

contact maps, spatial-distance distributions, and single-cell structural heterogeneity.

We will highlight applications across human cell lines, different organisms, and stages

of the cell cycle, as well as the potential to predict how disease-associated structural

variants perturb genome folding. By integrating interpretable physical constraints with

generative learning models, this approach provides a route toward mechanistic and

predictive modeling of three-dimensional genome organization.