Generating equilibrium molecules with deep neural networks
This work addresses the problem of molecular discovery for chemists and material scientists, but it appears incremental as it combines existing concepts from atomistic neural networks and auto-regressive generative models.
The authors tackled the challenge of discovering atomistic systems with desirable properties by introducing an autoregressive convolutional deep neural network that generates molecular equilibrium structures, demonstrating its capability to generate molecules close to equilibrium for constitutional isomers of C7O2H10.
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$_{10}$.