University of Cambridge > > Machine learning in Physics, Chemistry and Materials discussion group (MLDG) > Reinforcement Learning for 3D Molecular Design

Reinforcement Learning for 3D Molecular Design

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Bingqing Cheng .

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 (3D) 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. 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. 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. 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.

[1] G. N. C. Simm, R. Pinsler, J. M. Hernández-Lobato in Proceedings of the 37th International Conference on Machine Learning, PMLR , 2020, pp. 8959–8969, [2] G. N. C. Simm, R. Pinsler, G. Csányi, J. M. Hernández-Lobato in 9th International Conference on Learning Representations, ICLR 2021 ,

This talk is part of the Machine learning in Physics, Chemistry and Materials discussion group (MLDG) series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2024, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity