Guiding Deep Molecular Optimization with Genetic Exploration

TL;DR

This paper introduces GEGL, a novel deep learning framework guided by genetic exploration for molecular design, significantly outperforming existing methods in optimizing molecular properties.

Contribution

The paper proposes a new genetic expert-guided learning framework that enhances deep neural network training for molecular generation, achieving state-of-the-art results.

Findings

GEGL achieves a score of 31.40 on octanol-water partition coefficient optimization.

GEGL attains the highest scores on 19 out of 20 GuacaMol benchmark tasks.

GEGL obtains perfect scores on three GuacaMol tasks.

Abstract

De novo molecular design attempts to search over the chemical space for molecules with the desired property. Recently, deep learning has gained considerable attention as a promising approach to solve the problem. In this paper, we propose genetic expert-guided learning (GEGL), a simple yet novel framework for training a deep neural network (DNN) to generate highly-rewarding molecules. Our main idea is to design a "genetic expert improvement" procedure, which generates high-quality targets for imitation learning of the DNN. Extensive experiments show that GEGL significantly improves over state-of-the-art methods. For example, GEGL manages to solve the penalized octanol-water partition coefficient optimization with a score of 31.40, while the best-known score in the literature is 27.22. Besides, for the GuacaMol benchmark with 20 tasks, our method achieves the highest score for 19 tasks, in comparison with state-of-the-art methods, and newly obtains the perfect score for three tasks.