Constrained Bayesian Optimization for Automatic Chemical Design
Ryan-Rhys Griffiths University of Cambridge、José Miguel Hernández-Lobato University of Cambridge Alan Turing Institute Microsoft Research
Abstract
Automatic Chemical Design is a framework for generating novel molecules with optimized properties. The original scheme, featuring Bayesian optimization over the latent space of a variational autoencoder, suffers from the pathology that it tends to produce invalid molecular structures. First, we demonstrate empirically that this pathology arises when the Bayesian optimization scheme queries latent points far away from the data on which the variational autoencoder has been trained. Secondly, by reformulating the search procedure as a constrained Bayesian optimization problem, we show that the effects of this pathology can be mitigated, yielding marked improvements in the validity of the generated molecules. We posit that constrained Bayesian optimization is a good approach for solving this class of training set mismatch in many generative tasks involving Bayesian optimization over the latent space of a variational autoencoder.
中文速览
自动化化学分子设计（Automatic Chemical Design）通过贝叶斯优化（Bayesian optimization）在变分自编码器（variational autoencoder, VAE）的潜空间中搜索性质优异的新分子，但原始方案存在一个严重缺陷：优化过程频繁探索远离训练数据的潜空间区域，导致解码出大量无效分子结构。本文首先通过诊断实验证实了这一"训练集不匹配"现象的根本原因，随后将搜索过程重新表述为带约束的贝叶斯优化问题，利用贝叶斯神经网络（Bayesian neural network）学习一个约束函数，强制优化只在能解码出有效分子的潜空间区域内进行。实验结果表明，带约束的贝叶斯优化使有效且类药物分子的生成比例从不足5%大幅提升至超过80%，显著优于原始基准方法。这一方案为所有基于VAE潜空间贝叶斯优化的生成式任务中普遍存在的训练集不匹配问题提供了一种通用的解决思路。
原文 arXiv:1709.05501;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1709.05501v6