Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

GraphDF: A Discrete Flow Model for Molecular Graph Generation

About

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.

Youzhi Luo, Keqiang Yan, Shuiwang Ji• 2021

Related benchmarks

TaskDatasetResultRank
Molecular GenerationZINC250K
Uniqueness99.97
68
Property optimizationZINC250k (test)
1st Order Metric0.844
33
Molecular GenerationQM9
Validity89
30
Constrained Property OptimizationZINC250K
Improvement5.62
27
Molecular GenerationZINC 250K (train/test)
Uniqueness0.9997
12
Graph generationEgo-small (test)
Degree0.04
11
Distribution-learningZINC250K
Uniqueness99.8
10
Distribution-learningQM9
Uniqueness67.2
10
Molecular GenerationQM9 (train test)
Uniqueness99.2
10
Molecular GenerationZINC250K MOSES (test)
FCD0.512
10
Showing 10 of 16 rows

Other info

Follow for update