Chemical Engg. Seminar Series : Dr. Soumajit Dutta.

Integrating Physics-Based and Data-Driven Models for Molecular Innovation.

Understanding molecular processes at the mechanistic level is central to advancing molecular engineering solutions for pressing societal challenges. Physics-based molecular simulations, which numerically integrate the equations of motion, provide atomistic insight into the dynamics of molecular systems. Alongside these traditional approaches, recent advances in data-driven learning offer a complementary paradigm, where these methods uncover patterns in high dimensional data using backpropagation-based training.

In this seminar, I discuss how data-driven learning improves interpretability, accuracy, and computational efficiency of molecular simulation. Two case studies illustrate these ideas. First, Ishow how machine learning methods applied to high-dimensional molecular dynamics trajectories reveal differential allosteric signaling in proteins, offering insights for drug and biomolecule design. Second, I demonstrate how data-driven force fields and collective variables enable accurate and efficient modeling of defect dynamics in silicon carbide, a material of increasing importance in electronics, energy, and quantum technologies.

By bridging physics-based simulation with data-driven methods, we can accelerate discovery, improve design principles, and open new opportunities for chemical and materials engineering.