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.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2604.19562 [cs.LG] |
| (or arXiv:2604.19562v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2604.19562 arXiv-issued DOI via DataCite (pending registration) |
From: Carles Navarro [view email]
[v1]
Tue, 21 Apr 2026 15:13:41 UTC (2,995 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。