面向知识发现的药物ADMET情报预测方法  

Drug ADMET Intelligence Prediction Method for Knowledge Discovery

作　　者：郭勇 罗敏[1] 幸芮 GUO Yong;LUO Min;XING Rui(Business School,Central South University,Changsha 410083,China)

机构地区：[1]中南大学商学院,长沙410083

出　　处：《情报科学》2023年第2期95-100,156,共7页Information Science

摘　　要：【目的/意义】挖掘药物筛选工作中的隐性知识,借助机器学习的预测能力替代生物实验方法,减少制药流程的研发时间和经济成本。【方法/过程】提出一种面向知识发现的ADMET情报预测理论框架,以4种传统机器学习方法和2种集成学习方法,分别构建6种分类预测模型,提取药物的隐性知识,比较不同模型的优越性,评估最优模型的经济价值。【结果/结论】以药物分子描述符信息预测ADMET具有可行性,6种模型性能表现综合排序结果为随机森林、梯度提升决策树、Logistic回归、支持向量机、K近邻、高斯朴素贝叶斯。前沿信息技术能够有效应用于药物知识发现,信息经济学分析可预见创造可观收益,是未来制药工艺降本增效的重要手段。【创新/局限】未来应融合专家知识、追加试验验证、丰富参考指标。【Purpose/significance】Excavate the tacit knowledge in the screening of medicines, replace biological experimental methods with the prediction ability of machine learning(ML), and reduce R&D period and economic cost of pharmaceutical process.【Method/process】This paper proposes ADMET intelligence prediction theoretical framework for knowledge discovery and four traditional machine learning methods and two ensemble learning methods are used to construct Six classification prediction models. We extract the tacit knowledge, compare the advantages of different models, and evaluate the economic value of the optimal model.【Result/conclusion】It is feasible to predict ADMET with the molecular descriptor information. The comprehensive ranking results of the six models are RF, GBDT, LR, SVM, KNN and GNB. The cutting-edge information technologies can be effectively applied to the drug knowledge discovery. Information economics analysis can predict a positive revenue, which is an important means to reduce costs and increase efficiency of pharmaceutical processes in the future.【Innovation/limitation】We should integrate expert knowledge, add experimental verification and enrich reference indicators later.

关 键 词：机器学习 情报预测 知识发现 ADMET 经济价值 

分 类 号：G252.0[文化科学—图书馆学]