Efficient learning of non-autoregressive graph variational autoencoders for molecular graph generation

Youngchun Kwon,Dongseon Lee,Seokho Kang,Jiho Yoo,Won-Joon Son,Youn-Suk Choi

doi:10.1186/s13321-019-0396-x

Youngchun Kwon, Dongseon Lee + Show 4 more

Open Access

https://doi.org/10.1186/s13321-019-0396-x

Copy DOI

Abstract

With the advancements in deep learning, deep generative models combined with graph neural networks have been successfully employed for data-driven molecular graph generation. Early methods based on the non-autoregressive approach have been effective in generating molecular graphs quickly and efficiently but have suffered from low performance. In this paper, we present an improved learning method involving a graph variational autoencoder for efficient molecular graph generation in a non-autoregressive manner. We introduce three additional learning objectives and incorporate them into the training of the model: approximate graph matching, reinforcement learning, and auxiliary property prediction. We demonstrate the effectiveness of the proposed method by evaluating it for molecular graph generation tasks using QM9 and ZINC datasets. The model generates molecular graphs with high chemical validity and diversity compared with existing non-autoregressive methods. It can also conditionally generate molecular graphs satisfying various target conditions.

Highlights

In recent years, machine learning has been actively adopted to accelerate the discovery of novel molecules with desired properties [1,2,3,4]
To make the learning more efficient, we introduce an approximate graph matching procedure, which aims to alleviate the computational burden for the reconstruction loss
While the non-autoregressive approach has the advantage of fast and computationally efficient generation of molecular graphs without any iterative procedure, the existing methods suffer from low performance

Summary

Introduction

Machine learning has been actively adopted to accelerate the discovery of novel molecules with desired properties [1,2,3,4] While traditional approaches, such as manual design and enumeration [5,6,7,8], depend highly on domain knowledge and intuition of human experts, machine learning approaches have allowed automated design of desired molecules in a data-driven manner. They have attracted considerable attention from academia and industry.

Methods

Results

Conclusion