Traversing Chemical Space with Latent Potential Flows

Abstract

We consider the latent traversal problem in studying and exploring the chemical space with the learned latent space of a generative model. We propose a new framework, ChemFlow, that unified previous molecule manipulation and optimization method with a dynamical system perspective. Specifically, we formulate the problem as learning a vector field that transports the mass of the molecular distribution to the region with desired molecular properties or structure diversity. We also propose several alternative dynamics which exhibit various advantages over previous methods. We validate the efficacy of our proposed methods on both supervised and unsupervised molecule manipulation and optimization scenarios.