Deep Reinforcement Learning for De Novo Drug Design

• Alexander Tropsha  |  Mariya Popova  |  Olexandr Isayev

• UNC Eshelman School of Pharmacy  |  University of North Carolina USA

This model is an implementation of a novel computational strategy for de novo design of molecules with desired properties termed ReLeaSE (Reinforcement Learning for Structural Evolution).The most distinct innovative aspects of the approach proposed herein include the simple representation of molecules by their simplified molecular-input line-entry system (SMILES) strings only for both generative and predictive phases of the method and integration of these phases into a single workflow that includes a RL module. In the first phase of the method, generative and predictive models are trained separately with a supervised learning algorithm. In the second phase, both models are trained jointly with the RL approach to bias the generation of new chemical structures toward those with the desired physical and/or biological properties.In this system, the generative model is used to produce novel chemically feasible molecules, that is, it plays a role of an agent, whereas the predictive model (that predicts the properties of novel compounds) plays the role of a critic, which estimates the agent’s behavior by assigning a numerical reward (or penalty) to every generated molecule. The reward is a function of the numerical property generated by the predictive model, and the generative model is trained to maximize the expected reward.

Input variables : Simple representation of molecules by their simplified molecular-input line-entry system strings
Output Variables : Chemical compounds with desired physical, chemical, and/or bioactivity properties

Visit Model : advances.sciencemag.org

Additional links : advances.sciencemag.org