Molecular Generative Adversarial Network with Multi-Property Optimization

2024-03-29

arXiv.org

AI Summary

To address training instability, high computational cost, and poor scalability to large-scale chemical libraries in Monte Carlo Tree Search (MCTS)-based GAN approaches for molecular generation in drug discovery, this paper proposes InstGAN. The framework integrates token-level molecular representation, generative adversarial networks (GANs), and actor-critic reinforcement learning. It introduces a novel “instantaneous + global” dual-scale reward mechanism and incorporates a maximum-entropy policy to mitigate mode collapse. Optimizing multiple properties jointly—including drug-likeness, aqueous solubility, and target binding affinity—InstGAN achieves significant improvements over state-of-the-art GAN and RL baselines on ZINC and MOSES benchmarks. It matches or exceeds current SOTA methods in diversity, validity, and drug-likeness while substantially improving inference efficiency and enabling scalable deployment over billion-compound libraries.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Deep generative models, such as generative adversarial networks (GANs), have been employed for $de~novo$ molecular generation in drug discovery. Most prior studies have utilized reinforcement learning (RL) algorithms, particularly Monte Carlo tree search (MCTS), to handle the discrete nature of molecular representations in GANs. However, due to the inherent instability in training GANs and RL models, along with the high computational cost associated with MCTS sampling, MCTS RL-based GANs struggle to scale to large chemical databases. To tackle these challenges, this study introduces a novel GAN based on actor-critic RL with instant and global rewards, called InstGAN, to generate molecules at the token-level with multi-property optimization. Furthermore, maximized information entropy is leveraged to alleviate the mode collapse. The experimental results demonstrate that InstGAN outperforms other baselines, achieves comparable performance to state-of-the-art models, and efficiently generates molecules with multi-property optimization. The source code will be released upon acceptance of the paper.

Problem

• Overcoming instability in GAN and RL training for molecular generation
• Reducing high computational costs of MCTS sampling in drug discovery
• Achieving multi-property optimization in token-level molecule generation

Innovation

• Actor-critic RL with instant rewards
• Token-level molecular generation optimization
• Maximized information entropy prevents mode collapse