Generating equilibrium molecules with deep neural networks

TL;DR

This paper presents a new deep neural network architecture that generates equilibrium molecular structures by sequentially placing atoms, combining atomistic neural networks with autoregressive generative models.

Contribution

It introduces a novel autoregressive convolutional neural network for generating 3D molecular configurations based on atomic distances and charges.

Findings

Successfully generates molecules close to equilibrium structures.

Capable of modeling constitutional isomers of C7O2H10.

Integrates concepts from image and speech generative models.

Abstract

Discovery of atomistic systems with desirable properties is a major challenge in chemistry and material science. Here we introduce a novel, autoregressive, convolutional deep neural network architecture that generates molecular equilibrium structures by sequentially placing atoms in three-dimensional space. The model estimates the joint probability over molecular configurations with tractable conditional probabilities which only depend on distances between atoms and their nuclear charges. It combines concepts from state-of-the-art atomistic neural networks with auto-regressive generative models for images and speech. We demonstrate that the architecture is capable of generating molecules close to equilibrium for constitutional isomers of C$_7$O$_2$H$_{10}$.