基于深度神经网络对抗乳腺癌候选药物ERα生物活性的预测  

Prediction of biological activity of breast cancer candidate ERαbased on deep neural network

作　　者：花蕊 朱家明[1] HUA Rui;ZHU Jia-ming(School of Statistics and Applied Mathematics,Anhui University of Finance and Economics,Bengbu 233041,China)

机构地区：[1]安徽财经大学统计与应用数学学院,安徽蚌埠233041

出　　处：《陕西理工大学学报（自然科学版）》2023年第2期47-53,共7页Journal of Shaanxi University of Technology:Natural Science Edition

基　　金：国家社会科学基金项目(22BTJ048);安徽财经大学研究生科研创新基金项目(ACYC2021404)。

摘　　要：针对抗乳腺癌候选药物的ERα生物活性建立定量结构-活性关系模型预测药物化合物的生物活性,首先通过主成分分析法,以化合物的生物活性值(IC_(50))为因变量,729个分子描述符为自变量。再利用皮尔逊相关系数,剔除高度相关的变量,最终得到20个最具有显著影响的变量。使用支持向量机(SVM)和深度神经网络(DNN)两种模型分别建立化合物对ERα生物活性定量预测模型,利用评价指标对两模型结果进行比较。结果显示DNN模型的预测结果较好,其均方根误差为0.7355,均方误差为0.5354,平均绝对百分比误差为0.0861。研究得出的预测模型可以极大地节省药物研发时间,为新型抗乳腺癌先导化合物的药物研究提供实验和理论支持。In order to study the ERαbioactivity of breast cancer candidate,a quantitative structure-activity relationship(QSAR)model was established by Principal Component Analysis with the biological activity value(IC_(50))of the compounds as the dependent variable and 729 molecular descriptors as the independent variable.Using the Pearson product-moment Correlation Coefficient,we eliminated the highly correlated variables and came up with the 20 with the most significant effects.The quantitative prediction model of ERαbioactivity was established by using the data package of R language,Support Vector Machine(SVM)and Deep Neural Network(DNN),and the results were compared by using evaluation index.The result of DNN model is good,the root mean square error is 0.7355,the average absolute error is 0.5354,and the average absolute error is 0.0861.The results show that the time and cost of drug development can be greatly saved by computer-aided analysis and prediction of the biological activity of the compounds against ERα.

关 键 词：乳腺癌药物 ERα生物活性 主成分分析法 支持向量机 深度神经网络 预测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] R965.1[自动化与计算机技术—计算机科学与技术]