GraphDF: A Discrete Flow Model for Molecular Graph Generation

Youzhi Luo, Keqiang Yan, Shuiwang Ji
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:7192-7203, 2021.

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-luo21a, title = {GraphDF: A Discrete Flow Model for Molecular Graph Generation}, author = {Luo, Youzhi and Yan, Keqiang and Ji, Shuiwang}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {7192--7203}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/luo21a/luo21a.pdf}, url = {https://proceedings.mlr.press/v139/luo21a.html}, 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.} }
Endnote
%0 Conference Paper %T GraphDF: A Discrete Flow Model for Molecular Graph Generation %A Youzhi Luo %A Keqiang Yan %A Shuiwang Ji %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-luo21a %I PMLR %P 7192--7203 %U https://proceedings.mlr.press/v139/luo21a.html %V 139 %X 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.
APA
Luo, Y., Yan, K. & Ji, S.. (2021). GraphDF: A Discrete Flow Model for Molecular Graph Generation. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:7192-7203 Available from https://proceedings.mlr.press/v139/luo21a.html.

Related Material