A HIERARCHICAL FRAGMENT-BASED MODEL FOR 3D DRUG-LIKE MOLECULE GENERATION

Bo Qiang, Yuxuan Song, Minkai Xu, Jingjing Gong, Hao Zhou, Wei-Ying Ma, Yanyan Lan

Keywords: Drug Design, Molecule Generation, Deep Learning, Computational Biology

TL;DR: This paper introduced a hierarchical generative model for fragment-based drug-like molecule generation.

Abstract: De novo design of hit molecules is an important task in drug discovery. With the help of deep generative models, 3D molecular point set generation for smaller molecules (QM9) has been proposed by a few researchers. However, it is a non-trivial task to generate drug-like molecules which have relatively large atom numbers in the 3D space. Inspired by the human prior from domain experts, we propose a hierarchical fragment-based model. In order to avoid fragment collisions and maintain chemical validity, we solve the problem by generating high-level features and then sampling specific fragments and edges conditioned on the former. This hierarchical framework can capture basic chemical rules while generating 3D molecules of high quality. To evaluate our model's ability to sample molecules from the drug-like chemical space, we tested our method on multiple metrics. Among all evaluated metrics, our model outperforms the baseline model by a large margin.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Machine Learning for Sciences (eg biology, physics, health sciences, social sciences, climate/sustainability )