AI augmented molecular simulations for predicting protein and RNA structural ensembles
AI is now everywhere in chemistry, from structure prediction to molecule generation to automated synthesis. The excitement is real, but so is the unease about what is genuinely predictive and what is closer to impressive memorization. In this colloquium I will take a statistical physicist’s perspective and use examples from my group’s work to argue for cautious, but clear, enthusiasm for AI in chemistry and allied fields. I will show how we combine generative AI with statistical mechanics to learn Boltzmann weighted ensembles from limited training data, and then extrapolate across temperature, pressure, and other thermodynamic conditions reducing the need for explicit, expensive simulations or experiments. I will highlight the breadth of these methods through applications that include nucleation of crystal polymorphs under nanoconfinement, prediction of protein and RNA structural ensembles, and conformation selective drug discovery efforts aimed at Alzheimer’s disease and hypertension. Time permitting, I will discuss briefly what I think are the biggest challenges facing chemistry research and education as we proceed with the perhaps inevitable adoption of AI.
Please use this link to access Zoom.
For any questions, please contact Hyacinth Camillieri at [email protected]