Chen's group
Modeling Biomolecule Conformations with Enhanced Sampling and Deep Learning
Predicting biomolecular conformations is crucial but standard all-atom MD is too costly for large systems and long timescales. We tackle this with enhanced-sampling methods on collective-variable free-energy landscapes and further boost efficiency by integrating AI. Building on recent deep-learning advances that generate Boltzmann-consistent structures, we address the key challenge of modeling “constrained ensembles” under experimental, environmental, or structural conditions. Our AI frameworks—spanning protein-language-model embeddings and physics-guided diffusion models—produce protein conformations conditioned on data, interactions with material surfaces, or constraints from organic linkers, enabling accurate, generalizable ensemble generation.
Stochastic Electronic Structure Theory
We delve into the realm of ab initio methods, pivotal in uncovering the properties of molecules and materials. These methods, aiming to solve intricate Schrödinger equations, encounter challenges for larger systems due to computational demands. Our focus lies in advancing stochastic electronic structure techniques, leveraging random vectors to create a novel low-rank approximation. By doing so, we achieve linear scaling for Density Functional Theory (DFT) and TD-DFT, holding the promise to model complex materials like nanostructures and interfaces. This innovation has the potential to expand computational materials science horizons, bridging the gap for sizable experimental systems.
Thermodynamics with strong light-matter interaction
The historical interplay between light and matter, spanning molecules to condensed matter, is harnessed through techniques like spectroscopy across disciplines. While often theoretically approached with "weak coupling," allowing perturbative study, pronounced light-matter interactions occur in optical cavities, necessitating integrated modeling. Our research delves into understanding strong light-matter interactions, altering ground state properties in multi-molecule systems. Using machine learning, we model these interactions' effects on intermolecular dynamics, coupled with molecular dynamics simulations revealing structural and thermodynamic insights within optical cavities.
