Machine learning for molecular geometry optimization and 3D structure generation

Abstract

Artificial intelligence (AI) has infiltrated all fields of science, from high-energy particle newlinephysics to biology to computational chemistry. In the last couple of decades, there has been newlinetremendous advancement in machine learning (ML) applications in computational chemistry. newlineDeep learning (DL) has achieved some success in the automation of feature design, physicochemical newlineproperty prediction, accelerated chemical space search, and the design of new drug-like newlinemolecules. Much work is still needed in terms of property prediction of inorganic molecules, newlinealong with the search and design of new molecules and material design with desired properties. newlineThis research aims to develop machine learning methods for 3D structure generation and newlinemolecular geometry optimization. Use of neural network potential (NNPs) can accelerate the newlineprocess of 3D structure generation and molecular geometry optimization. Various neural network newlinepotentials (NNPs) have been reported in the literature to be as fast as force fields and newlineas accurate as DFT. There has been a lack of standard comparative evaluation of these NNPs, newlinewhich motivated us to do a benchmark study on NNPs. In this benchmark study, we evaluate newlineand compare four NNPs, i.e., ANI, PhysNet, SchNet, and BAND-NN, for their accuracy in newlineenergy prediction, transferability to larger molecules, ability to produce accurate PES, and applicability newlinein geometry optimization. In the context of 3D structure generation (Molecules and newlinematerial design), there are two major components: search algorithm and property predictor. newlineWe need a fast and accurate method to predict the energy of the given system to accelerate newlinethe search in conformational space. For this, we developed a model known as DART, which newlinepredicts the energy of Gallium clusters using a Topological Atomistic Descriptor (TAD). TAD newlineis a very simple and elegant descriptor that tries to encode structural information by dividing newlinethe connectivity information using distance cutoffs. We show the DART models ability to newlinepredict the energies of Gallium c