Description
The goal of molecular crystal structure prediction (CSP) is to find all plausible polymorphs for a given molecule. This requires performing multimodal global optimization over a high dimensional search space. To tackle this challenge, we have developed open-source, and massively parallel genetic algorithm package in Python called GAtor with fitness, selection, crossover, and mutation operators specifically designed for CSP, but GAtor’s modularity allows our framework to be used for other purposes. We use descriptors for structural similarity and Scikit-learn’s Affinity Propagation and K-Means clustering algorithms in a couple of different ways to effect niching and overcome local minimum entrapment.Presenter(s): Speaker: Timothy Rose, Carnegie Mellon University Author: Farren Curtis, Carnegie Mellon University Author: Xiayue Li, Google Author: Noa Marom, Carnegie Mellon University