TY - JOUR
AU - Yinjun Jia
AU - Bowen Gao
AU - Jiaxin Tan
AU - Jiqing Zheng
AU - Xin Hong
AU - Wenyu Zhu
AU - Haichuan Tan
AU - Yuan Xiao
AU - Liping Tan
AU - Hongyi Cai
AU - Yanwen Huang
AU - Zhiheng Deng
AU - Xiangwei Wu
AU - Yue Jin
AU - Yafei Yuan
AU - Jiekang Tian
AU - Wei He
AU - Weiying Ma
AU - Yaqin Zhang
AU - Lei Liu
AU - Chuangye Yan
AU - Wei Zhang
AU - Yanyan Lan
AB - Recent breakthroughs in protein structure prediction have opened new avenues for genome-wide drug discovery, yet existing virtual screening methods remain computationally prohibitive. We present DrugCLIP, a contrastive learning framework that achieves ultrafast and accurate virtual screening, up to 10 million times faster than docking, while consistently outperforming various baselines on in silico benchmarks. In wet-lab validations, DrugCLIP achieved a 15% hit rate for norepinephrine transporter, and structures of two identified inhibitors were determined in complex with the target protein. For thyroid hormone receptor interactor 12, a target that lacks holo structures and small-molecule binders, DrugCLIP achieved a 17.5% hit rate using only AlphaFold2-predicted structures. Finally, we released GenomeScreenDB, an open-access database providing precomputed results for ~10,000 human proteins screened against 500 million compounds, pioneering a drug discovery paradigm in the post-AlphaFold era.
BT - Science
DA - 2026-01-08
DO - 10.1126/science.ads9530
IS - 6781
N2 - Recent breakthroughs in protein structure prediction have opened new avenues for genome-wide drug discovery, yet existing virtual screening methods remain computationally prohibitive. We present DrugCLIP, a contrastive learning framework that achieves ultrafast and accurate virtual screening, up to 10 million times faster than docking, while consistently outperforming various baselines on in silico benchmarks. In wet-lab validations, DrugCLIP achieved a 15% hit rate for norepinephrine transporter, and structures of two identified inhibitors were determined in complex with the target protein. For thyroid hormone receptor interactor 12, a target that lacks holo structures and small-molecule binders, DrugCLIP achieved a 17.5% hit rate using only AlphaFold2-predicted structures. Finally, we released GenomeScreenDB, an open-access database providing precomputed results for ~10,000 human proteins screened against 500 million compounds, pioneering a drug discovery paradigm in the post-AlphaFold era.
PY - 2026
EP - eads9530
T2 - Science
TI - Deep contrastive learning enables genome-wide virtual screening
UR - https://www.science.org/doi/10.1126/science.ads9530
VL - 391
Y2 - 2026-01-12
ER -