Coarse-grained models enable large-scale simulations of complex biomolecular systems over long timescales. We have introduced bottom-up approaches based on contrastive learning to parameterize these models using atomistic simulations, and top-down approaches based on maximum entropy optimization to parameterize them from experimental data. We performed coarse-grained simulations to study the liquid-liquid phase separation of biomolecules, which has emerged as a fundamental principle for organizing various cellular structures. In addition, we simulate the 3D conformations of the human genome to understand its organizational principles and the role of chromatin structure in gene regulation. Our studies help elucidate the molecular determinants for the stability, internal organization, and coarsening of biocondensates, and the contribution of various condensates to genome organization.
