(710b) Strategies and Software for Accelerating Inorganic Molecular Design

Chemical space is vast, with best estimates suggesting we have as yet characterized a tiny fraction of all possible compounds. The need for efficient discovery of new inorganic molecules as building blocks for materials and catalysts mandates that we identify smart ways to map out and explore transition metal chemical space. Virtual high throughput screening, typically driven by first-principles, density functional theory (DFT) calculations, has emerged as a powerful tool for the discovery of new materials. However, approximate DFT is simultaneously too computationally costly for full enumeration of chemical space and also too inaccurate for robust prediction of properties with correlated electrons. I will describe our recent development of the first open source toolkit for inorganic molecular discovery, molSimplify, that aims to address both these challenges. i) We introduce a divide-and-conquer approach to generate precise geometries of new inorganic complexes using building blocks from multi-million molecule organic libraries for automated simulation of diverse compounds with DFT. ii) We train and incorporate a neural network to predict quantum mechanical properties in microseconds for use in efficient screening workflows as opposed to minutes or hours with DFT. iii) We develop automated strategies to uncover structure-property correlations for molecular design. iv) We reveal structure-method-sensitivity relationships that guide when going beyond approximate-DFT is essential. v) We incorporate all these advances into evolutionary algorithm workflows in the molSimplify automated design (mAD) toolkit that adapt on the fly from data-driven models to DFT and beyond for robust prediction. I will review how implementing all of these techniques in our open source toolkit has advanced applications in catalysis and materials.