Target-aware Variational Auto-encoders for Ligand Generation with Multimodal Protein Representation Learning

TL;DR

This paper introduces TargetVAE, a novel target-aware variational auto-encoder that leverages a multimodal protein representation to generate high-affinity ligands for arbitrary targets, streamlining drug discovery.

Contribution

It unifies multiple protein representations into a single model using graph Transformers, enabling ligand generation without tailored networks for each protein.

Findings

Outperforms existing methods in ligand generation quality.

Effectively predicts binding affinities and docking scores.

Generates ligands for unseen protein targets with promising results.

Abstract

Without knowledge of specific pockets, generating ligands based on the global structure of a protein target plays a crucial role in drug discovery as it helps reduce the search space for potential drug-like candidates in the pipeline. However, contemporary methods require optimizing tailored networks for each protein, which is arduous and costly. To address this issue, we introduce TargetVAE, a target-aware variational auto-encoder that generates ligands with high binding affinities to arbitrary protein targets, guided by a novel multimodal deep neural network built based on graph Transformers as the prior for the generative model. This is the first effort to unify different representations of proteins (e.g., sequence of amino-acids, 3D structure) into a single model that we name as Protein Multimodal Network (PMN). Our multimodal architecture learns from the entire protein structures and is able to capture their sequential, topological and geometrical information. We showcase the superiority of our approach by conducting extensive experiments and evaluations, including the assessment of generative model quality, ligand generation for unseen targets, docking score computation, and binding affinity prediction. Empirical results demonstrate the promising performance of our proposed approach. Our software package is publicly available at https://github.com/HySonLab/Ligand_Generation