Abstract: Data-driven modeling and machine learning have opened new paradigms and opportunities in the understanding, design, and simulation of soft and biological materials. In the first part of this talk, I will describe our use of coarse-grained molecular dynamics, graph matching, variational autoencoders, and Bayesian optimization to perform high-throughput virtual screening of the sequence space of π-conjugated oligopeptides. Direct screening of only a small fraction of the allowable sequence space allows us to identify a small number of optimized oligopeptide chemistries capable of spontaneous self-assembly into optically and electronically active biocompatible nanoaggregates.

In the second part of this talk, I will describe our development of a deep-learning approach to estimate slow collective variables from molecular simulation trajectories and the use of these coordinates to train highly efficient latent space molecular simulators. The high cost of all-atom calculations restricts molecular simulations to sub-millisecond timescales. By learning the slowest dynamical modes governing the long-time dynamics from short molecular dynamics simulations, we have trained autonomous dynamical propagators within low-dimensional latent spaces together with generative models of the all-atom configurations to synthesize thermodynamically and dynamically correct simulation trajectories at ~10⁶ lower cost than conventional molecular dynamics calculations.

Bio: Andrew Ferguson is an associate professor at the Pritzker School of Molecular Engineering at the University of Chicago. He received an M.Eng. in chemical engineering from Imperial College London and a Ph.D. in chemical and biological engineering from Princeton University. His research uses theory, simulation, and machine learning to understand and design self-assembling materials, macromolecular folding, and antiviral therapies.