Application of Generative Autoencoder in <i>De Novo</i> Molecular Design

TLDR

Generating novel molecules with desirable pharmacological and physicochemical properties remains a major challenge in computational chemistry. This study investigates using autoencoders, a deep learning approach, for de novo molecular design. Various generative autoencoders mapped molecular structures into a continuous latent space and back, and their performance as structure generators was evaluated. The latent space preserves chemical similarity, enabling generation of analogue structures, and systematic searches produced novel compounds predicted to target dopamine receptor type 2 and identified actives absent from the training set.

Abstract

Abstract A major challenge in computational chemistry is the generation of novel molecular structures with desirable pharmacological and physiochemical properties. In this work, we investigate the potential use of autoencoder, a deep learning methodology, for de novo molecular design. Various generative autoencoders were used to map molecule structures into a continuous latent space and vice versa and their performance as structure generator was assessed. Our results show that the latent space preserves chemical similarity principle and thus can be used for the generation of analogue structures. Furthermore, the latent space created by autoencoders were searched systematically to generate novel compounds with predicted activity against dopamine receptor type 2 and compounds similar to known active compounds not included in the trainings set were identified.