Structure-guided molecular design with contrastive 3D protein-ligand learning

TL;DR

This paper introduces a unified structure-guided molecular design framework combining contrastive 3D protein-ligand encoding with autoregressive generation, enabling targeted drug candidate discovery.

Contribution

It presents an SE(3)-equivariant transformer for contrastive 3D structure encoding and integrates it into a multimodal model for targeted molecule generation.

Findings

Achieves competitive zero-shot virtual screening results.

Generates target-specific molecules conditioned on protein pockets or ligands.

Produces candidates with favorable binding properties across diverse targets.

Abstract

Structure-based drug discovery faces the dual challenge of accurately capturing 3D protein-ligand interactions while navigating ultra-large chemical spaces to identify synthetically accessible candidates. In this work, we present a unified framework that addresses these challenges by combining contrastive 3D structure encoding with autoregressive molecular generation conditioned on commercial compound spaces. First, we introduce an SE(3)-equivariant transformer that encodes ligand and pocket structures into a shared embedding space via contrastive learning, achieving competitive results in zero-shot virtual screening. Second, we integrate these embeddings into a multimodal Chemical Language Model (MCLM). The model generates target-specific molecules conditioned on either pocket or ligand structures, with a learned dataset token that steers the output toward targeted chemical spaces, yielding candidates with favorable predicted binding properties across diverse targets.