In computation chemistry, we utilize computer-based methods to approximate solutions to the Schrodinger equation. These methods are successful at giving chemical insight, however they can be quite costly, as well as limiting, depending on the accuracy you want to achieve and the size of your system. Using Behler and Parrinello’s high dimensional atomic neural network potentials (NNP) and a modified form of their symmetry functions, we have developed the ANAKIN-ME (Accurate NeurAl networK engINe for Molecular Energies), or ANI, methodology, which bridges this gap between cost and accuracy. ANI is a transferable machine learning potential for organic molecules, successful at reproducing the chemical accuracy of its reference level of theory. The ANI methodology supplies accurate predictions of QM reference calculations while balancing computational cost comparable to classical force fields, providing chemical insight and understanding into the energetics and dynamics of molecular systems for the purpose of studying materials and drug design.<br/>As an effort to extend the ANI methodology, I am currently working on the development of a NNP that will accurately predict the energy of the system, as well as energy-based atomic electrostatic parameters. Because the atomic energy and charge values will directly correspond with one another, this will allow for the charge transfer models of conceptual DFT to be tested. Further chemical insight can be gained by estimating other parameters, such as electronegativity and hardness. The ANIAIM partition could also allow for prediction of reactivity descriptors, such as fukui functions, while providing this data at a fraction of the cost.

ANI-AIM: Using machine learning for energy-based electrostatics

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.