Utilizing Reinforcement Learning for de novo Drug Design

TL;DR

This paper presents a comprehensive framework using reinforcement learning to generate novel drug molecules with desired activity, emphasizing the importance of diversity and stability in the generation process.

Contribution

It systematically compares various on- and off-policy reinforcement learning algorithms and replay buffers for de novo drug design, providing an open-source tool for future research.

Findings

Using both high- and low-scoring molecules improves structural diversity.

Including all generated molecules enhances stability of on-policy algorithms.

Off-policy algorithms can increase diversity and active molecule count, but may require longer exploration.

Abstract

Deep learning-based approaches for generating novel drug molecules with specific properties have gained a lot of interest in the last few years. Recent studies have demonstrated promising performance for string-based generation of novel molecules utilizing reinforcement learning. In this paper, we develop a unified framework for using reinforcement learning for de novo drug design, wherein we systematically study various on- and off-policy reinforcement learning algorithms and replay buffers to learn an RNN-based policy to generate novel molecules predicted to be active against the dopamine receptor DRD2. Our findings suggest that it is advantageous to use at least both top-scoring and low-scoring molecules for updating the policy when structural diversity is essential. Using all generated molecules at an iteration seems to enhance performance stability for on-policy algorithms. In addition, when replaying high, intermediate, and low-scoring molecules, off-policy algorithms display the potential of improving the structural diversity and number of active molecules generated, but possibly at the cost of a longer exploration phase. Our work provides an open-source framework enabling researchers to investigate various reinforcement learning methods for de novo drug design.