Molecular Attributes Transfer from Non-Parallel Data

TL;DR

This paper introduces a novel style transfer approach for molecular optimization that learns from non-parallel data, enabling property enhancement without needing predefined attribute functions or paired datasets.

Contribution

It formulates molecular optimization as a style transfer problem and develops a generative model that leverages adversarial training, autoencoders, and flow techniques to improve molecules.

Findings

Outperforms state-of-the-art methods in toxicity modification.

Enhances synthesizability of molecules effectively.

Successfully preserves molecular content during optimization.

Abstract

Optimizing chemical molecules for desired properties lies at the core of drug development. Despite initial successes made by deep generative models and reinforcement learning methods, these methods were mostly limited by the requirement of predefined attribute functions or parallel data with manually pre-compiled pairs of original and optimized molecules. In this paper, for the first time, we formulate molecular optimization as a style transfer problem and present a novel generative model that could automatically learn internal differences between two groups of non-parallel data through adversarial training strategies. Our model further enables both preservation of molecular contents and optimization of molecular properties through combining auxiliary guided-variational autoencoders and generative flow techniques. Experiments on two molecular optimization tasks, toxicity modification and synthesizability improvement, demonstrate that our model significantly outperforms several state-of-the-art methods.