GraphDF: A Discrete Flow Model for Molecular Graph Generation

TL;DR

GraphDF introduces a discrete flow model for molecular graph generation, effectively modeling discrete structures with improved accuracy and efficiency over prior continuous latent variable approaches.

Contribution

The paper presents GraphDF, a novel discrete latent variable model based on normalizing flows, specifically designed for molecular graph generation, reducing computational costs and improving modeling accuracy.

Findings

Outperforms prior methods in random generation tasks

Achieves better property optimization results

Excels in constrained optimization scenarios

Abstract

We consider the problem of molecular graph generation using deep models. While graphs are discrete, most existing methods use continuous latent variables, resulting in inaccurate modeling of discrete graph structures. In this work, we propose GraphDF, a novel discrete latent variable model for molecular graph generation based on normalizing flow methods. GraphDF uses invertible modulo shift transforms to map discrete latent variables to graph nodes and edges. We show that the use of discrete latent variables reduces computational costs and eliminates the negative effect of dequantization. Comprehensive experimental results show that GraphDF outperforms prior methods on random generation, property optimization, and constrained optimization tasks.