Automating molecular design using deep reinforcement learning (RL) has the potential to greatly accelerate the search for novel materials. Despite recent progress on leveraging graph representations to design molecules, such methods are fundamentally limited by the lack of three-dimensional information. To address this, we present a novel RL formulation for molecular design in Cartesian coordinates, thereby extending the class of molecules that can be built.[1] Our reward function is directly based on fundamental physical properties such as the electronic energy, which we approximate via fast quantum-chemical methods. Further, we propose a novel actor-critic architecture that exploits the symmetries of the design process through a rotationally covariant state-action representation based on a spherical harmonics series expansion.[2] In our experiments, we show that our agent can efficiently learn to solve molecular-design tasks from scratch that are unattainable with graph-based approaches. We demonstrate that our agent can build stable organic and inorganic molecules and molecular clusters. To enable progress towards de-novo molecular design, we introduce MolGym, an RL environment comprising several challenging molecular design tasks along with baselines.[3]

Reinforcement learning for molecular design guided by quantum chemistry

ACS Fall 2021

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