Videos

Machine learning-enabled enhanced sampling in biomolecular simulation and data-driven design of self-assembling photonic crystals and optoelectonic π-conjugated oligopeptides

Presenter
September 9, 2019
Abstract
Andrew Ferguson - University of Chicago, Materials Science and Engineering Data-driven modeling and machine learning present powerful tools that are opening up new paradigms and opportunities in the understanding, discovery, and design of soft and biological materials. In the first part of this talk, I will describe the use of autoencoding artificial neural networks to discover high variance collective variables in protein folding. By passing these variables to powerful enhanced sampling techniques we perform simultaneous on-the-fly variable discovery and accelerated sampling of protein folding funnels starting from no knowledge of the underlying free energy landscape or key collective variables. In the second part of this talk, I will describe our applications of nonlinear manifold learning, variational autoencoders, evolutionary algorithms, and Bayesian optimization in two materials design problems. (1) Colloidal particles with tunable anisotropic interactions may be functionalized to spontaneously assemble photonic crystals with omnidirectional band gaps in the visible regime. We have developed an inverse design protocol termed "landscape engineering" to perform automated inverse design of self-assembling diamond and pyrochlore photonic crystals possessing complete bandgaps. (2) Synthetic p-conjugated oligopeptides can be designed to assemble biocompatible nanowires possessing engineered optical and electronic functionality. De novo design and Edisonian testing of molecular chemistries is frustrated by the astronomical size of chemical space. By combining molecular simulation and Bayesian optimization we developed an active-learning protocol to perform a guided search through sequence space to identify high-performing molecular candidates for experimental testing.
Supplementary Materials