Facing small and biased data dilemma in drug discovery with enhanced federated learning approaches

作     者：Zhaoping Xiong Ziqiang Cheng Xinyuan Lin Chi Xu Xiaohong Liu Dingyan Wang Xiaomin Luo Yong Zhang Hualiang Jiang Nan Qiao Mingyue Zheng Zhaoping Xiong;Ziqiang Cheng;Xinyuan Lin;Chi Xu;Xiaohong Liu;Dingyan Wang;Xiaomin Luo;Yong Zhang;Hualiang Jiang;Nan Qiao;Mingyue Zheng

作者机构：Shanghai Institute for Advanced Immunochemical Studiesand School of Life Science and TechnologyShanghaiTech UniversityShanghai 200031China Drug Discovery and Design CenterState Key Laboratory of Drug ResearchShanghai Institute of Materia MedicaChinese Academy of SciencesShanghai 201203China University of Chinese Academy of SciencesBeijing 100049China School of Information Science and TechnologyUniversity of Science and Technology of ChinaHefei 230000China Laboratory of Health IntelligenceHuawei Technologies Co.LtdShenzhen 518100China 

出 版 物：《Science China(Life Sciences)》 (中国科学（生命科学英文版）)

年 卷 期：2022年第65卷第3期

页      面：529-539页

核心收录：

学科分类：0710[理学-生物学] 0830[工学-环境科学与工程（可授工学、理学、农学学位）] 10[医学] 

基　　金：supported by the Shanghai Municipal Science and Technology Major Project the National Natural Science Foundation of China(81773634) the National Science and Technology Major Project of the Ministry of Science and Technology of China(2018ZX09711002) the Strategic Priority Research Program of the Chinese Academy of Sciences(XDA12050201 and XDA12020368)。 

主　　题：federated learning drug discovery Fed AMP Non-IID data 

摘      要：Artificial intelligence(AI)models usually require large amounts of high-quality training data,which is in striking contrast to the situation of small and biased data faced by current drug discovery pipelines.The concept of federated learning has been proposed to utilize distributed data from different sources without leaking sensitive information of the data.This emerging decentralized machine learning paradigm is expected to dramatically improve the success rate of AI-powered drug discovery.Here,we simulated the federated learning process with different property and activity datasets from different sources,among which overlapping molecules with high or low biases exist in the recorded values.Beyond the benefit of gaining more data,we also demonstrated that federated training has a regularization effect superior to centralized training on the pooled datasets with high biases.Moreover,different network architectures for clients and aggregation algorithms for coordinators have been compared on the performance of federated learning,where personalized federated learning shows promising results.Our work demonstrates the applicability of federated learning in predicting drug-related properties and highlights its promising role in addressing the small and biased data dilemma in drug discovery.