连接器
标杆管理
合理设计
化学
强化学习
计算机科学
三元运算
降级(电信)
耐受性
组合化学
发电机(电路理论)
人工智能
稳健性(进化)
计算生物学
生物系统
共价键
作者
Xunkai Yin,Yihui Cai,Shulan Mei,Songyun Xu,Hanyuan Zhang,Xiaoqian Peng,Wenyu Lu,Zhenzhen Zhu,Wenzhuo Xu,Zhe Zheng,Weijiang Lin,Yang Hu,Lihong Hu,Jian Liu
标识
DOI:10.1021/acs.jmedchem.6c02194
摘要
Abstract Proteolysis-targeting chimeras (PROTACs) represent a promising therapeutic strategy, yet rational linker design is limited by inadequate theoretical guidance. Herein, we report PotentialLinker, a novel reinforcement learning framework for PROTAC linker optimization that uses a normalized weighted harmonic mean (NWHM) score to predict degradation activity. Benchmarking on ternary structures confirms its performance without precomputed protein−protein complexes. In case studies of STAT3/RIPK1, it recapitulated structure−activity relationships (SAR) and recommended appropriate E3 ligases. Then, conducted by PotentialLinker, a potent FGFR4 degrader D11 was obtained (DC50 = 56.23 nM, Dmax = 89.58%). Due to its covalent interacting with the unique Cys552 in FGFR4, D11 possesses excellent FGFR4 selectivity. Furthermore, D11 also demonstrated favorable oral bioavailability (F = 21.3%) and potent antitumor efficacy (TGI = 87.07%) with acceptable tolerability in vivo. Taken together, this framework effectively bridges degradation theory and practical PROTAC design, which can be utilized for the rational design of PROTAC-based therapeutic agents.
科研通智能强力驱动
Strongly Powered by AbleSci AI