APS -APS March Meeting 2017 - Event - Analysing and Rationalising Molecular and Materials Databases Using Machine-Learning

摘要

Computational materials design promises to greatly accelerate the process of discovering new or more performant materials. Several collaborative efforts are contributing to this goal by building databasesof structures, containing between thousands and millionsof distinct hypothetical compounds, whose properties are computed by high-throughput electronic-structurecalculations. The complexity and sheer amount of information has made manual exploration, interpretation and maintenance of these databases a formidable challenge, making it necessary to resort to automatic analysis tools.Here we will demonstrate how, starting from a measure of (dis)similaritybetween database items built from a combination of local environment descriptors,it is possible to apply hierarchical clustering algorithms, as well as dimensionality reduction methods such as sketchmap, to analyse, classifyand interpret trends in molecular and materials databases, as well as to detect inconsistencies and errors. Thanks to the agnostic and flexible nature of theunderlying metric, we will show how our framework can be applied transparently to different kinds of systems ranging from organic molecules and oligopeptides to inorganic crystal structures as well as molecular crystals.