Reinforcement Learning for Molecular Design Guided by Quantum Mechanics

Gregor Simm, Robert Pinsler, Jose Miguel Hernandez-Lobato
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:8959-8969, 2020.

Abstract

Automating molecular design using deep reinforcement learning (RL) holds the promise of accelerating the discovery of new chemical compounds. Existing approaches work with molecular graphs and thus ignore the location of atoms in space, which restricts them to 1) generating single organic molecules and 2) heuristic reward functions. 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. Our reward function is directly based on fundamental physical properties such as the energy, which we approximate via fast quantum-chemical methods. To enable progress towards de-novo molecular design, we introduce MolGym, an RL environment comprising several challenging molecular design tasks along with baselines. In our experiments, we show that our agent can efficiently learn to solve these tasks from scratch by working in a translation and rotation invariant state-action space.

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-simm20b, title = {Reinforcement Learning for Molecular Design Guided by Quantum Mechanics}, author = {Simm, Gregor and Pinsler, Robert and Hernandez-Lobato, Jose Miguel}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {8959--8969}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/simm20b/simm20b.pdf}, url = {https://proceedings.mlr.press/v119/simm20b.html}, abstract = {Automating molecular design using deep reinforcement learning (RL) holds the promise of accelerating the discovery of new chemical compounds. Existing approaches work with molecular graphs and thus ignore the location of atoms in space, which restricts them to 1) generating single organic molecules and 2) heuristic reward functions. 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. Our reward function is directly based on fundamental physical properties such as the energy, which we approximate via fast quantum-chemical methods. To enable progress towards de-novo molecular design, we introduce MolGym, an RL environment comprising several challenging molecular design tasks along with baselines. In our experiments, we show that our agent can efficiently learn to solve these tasks from scratch by working in a translation and rotation invariant state-action space.} }
Endnote
%0 Conference Paper %T Reinforcement Learning for Molecular Design Guided by Quantum Mechanics %A Gregor Simm %A Robert Pinsler %A Jose Miguel Hernandez-Lobato %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-simm20b %I PMLR %P 8959--8969 %U https://proceedings.mlr.press/v119/simm20b.html %V 119 %X Automating molecular design using deep reinforcement learning (RL) holds the promise of accelerating the discovery of new chemical compounds. Existing approaches work with molecular graphs and thus ignore the location of atoms in space, which restricts them to 1) generating single organic molecules and 2) heuristic reward functions. 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. Our reward function is directly based on fundamental physical properties such as the energy, which we approximate via fast quantum-chemical methods. To enable progress towards de-novo molecular design, we introduce MolGym, an RL environment comprising several challenging molecular design tasks along with baselines. In our experiments, we show that our agent can efficiently learn to solve these tasks from scratch by working in a translation and rotation invariant state-action space.
APA
Simm, G., Pinsler, R. & Hernandez-Lobato, J.M.. (2020). Reinforcement Learning for Molecular Design Guided by Quantum Mechanics. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:8959-8969 Available from https://proceedings.mlr.press/v119/simm20b.html.

Related Material