子树权重等4种拓扑指标预测抗HIV病毒活性  

Prediction of anti-HIV activity using four topological indices including subtree weights

作　　者：杨雨 靳棒棒 李波[2] 张修梅 YANG Yu;JIN Bangbang;LI Bo;ZHANG Xiumei(College of Information Science and Engineering,Henan University Of Technology,Zhengzhou 450001,China;School of Software,Pingdingshan University,Pingdingshan 467000,Henan China;College of Statistics and Mathematics,Shanghai Lixin University Of Accounting and Finance,Shanghai 201620,China)

机构地区：[1]河南工业大学信息科学与工程学院,河南郑州450001 [2]平顶山学院软件学院,河南平顶山467000 [3]上海立信会计金融学院统计与数学学院,上海201620

出　　处：《华中科技大学学报（自然科学版）》2024年第11期78-84,共7页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：河南省科技厅国际科技合作资助项目(242102521023,232102521002);河南省科技厅科技攻关资助项目(232102210011)。

摘　　要：针对目前预测抗艾滋病毒(HIV)大多使用距离型指标导致模型泛化性差的问题,通过构造基于广义邻接矩阵的子树权重信息无丢失行列变换规则,实现对结构型指标子树权重指标的高效计算.同时结合Wiener指标、Harary指标和Schultz指标,利用机器学习经典监督学习算法(支持向量机(SVM)、K-近邻算法(KNN)和决策树算法)构建模型对化合物分子的抗HIV活性进行预测.实验结果表明:子树权重指标具有良好的特征区分能力和准确度(91.03%~99.61%),因此该指标可以作为一种有效的新药研发的新度量.Aiming at the issue of poor model generalization in current anti-human immunodeficiency virus(HIV)prediction methods that predominantly relied on distance-based indices,an efficient approach for computing structural subtree weight indices was proposed by constructing a lossless row transformation rule for subtree weight information based on a generalized adjacency matrix.By integrating Wiener,Harary,and Schultz indices,and utilizing classical supervised learning algorithms in machine learning(support vector machines(SVM),K-nearest neighbors(KNN)algorithm,decision trees algorithm),models were built to predict the anti-HIV activity of compound molecules.Experimental results show that subtree weight indices exhibit excellent feature discrimination and accuracy,ranging from 91.03%to 99.61%.Therefore,this index can serve as an effective new metric in new drug discovery.

关 键 词：子树权重指标 广义邻接矩阵 树与单双圈图 机器学习 新药研发 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]