面向抗乳腺癌候选药物拮抗雌激素受体α生物活性的定量构效关系模型构建  

Construction of quantitative structure-activity relationship model of biological of activity anti-breast cancer drug candidate estrogen receptor α

作　　者：吕婷婷 禹文韬 张慧琳[2] LYU Ting-ting;YU Wen-tao;ZHANG Hui-lin(School of Nursing,Hunan University of Chinese Medicine,Changsha 410208;The Second Xiangya Hospital of Central South University,Changsha 410011;School of Automation,Central South University,Changsha 410000)

机构地区：[1]湖南中医药大学护理学院,长沙410208 [2]中南大学湘雅二医院,长沙410011 [3]中南大学自动化学院,长沙410000

出　　处：《中南药学》2022年第11期2542-2548,共7页Central South Pharmacy

基　　金：湖南省卫生健康委2020年科研立项课题(No.20201035);2021年度创新型省份建设专项科普专题立项项目(No.2021ZK4180)。

摘　　要：目的 建立抗乳腺癌候选药物的分子描述符与该药物拮抗雌激素受体α(ERα)活性的定量构效关系(QSAR)预测模型,预测具有更好拮抗ERα活性的新药物分子或指导已有候选药物的结构优化。方法 使用某药物企业提供的乳腺癌ERα拮抗剂数据,首先通过改进的随机森林对化合物的分子结构描述符进行变量选择,选取前20个对生物活性具有显著影响的分子描述符;然后利用支持向量回归构建20个分子描述符对药物拮抗ERα活性的QSAR模型,再采用差分进化算法来优化模型的超参数。结果 数据经随机森林进行变量选择后从729个变量变为20个变量;对于药物拮抗ERα活性的基于支持向量回归的QSAR模型的拟合优度(R^(2))为0.7407,均方误差(MSE)为0.7034;而经过差分进化算法优化的QSAR模型对于药物拮抗ERα活性的R^(2)为0.935,MSE为0.1762。通过R^(2)和MSE的检验,优化后的模型具有较好的稳健性与较高的预测精度。结论 通过建立化合物分子描述符与其拮抗ERα活性的QSAR模型,找到对乳腺癌治疗药物的药物活性有重要影响的描述符,可用来指导乳腺癌新型治疗药物的设计。Objective To establish a quantitative structure-activity relationship (QSAR) prediction model of the molecular descriptor of anti-breast cancer drug candidate and the activity of the drug antagonistic estrogen receptor α (ERα),to predict new drug molecules with better ERα antagonistic activity or to optimize the structural of existing candidate drugs.Methods The ERα antagonistic data of breast cancer provided by a pharmaceutical company were used.The first 20 molecular descriptors with the most significant influence on the biological activity were selected by improved random forest,and then the QSAR model of 20 molecular descriptors of drug antagonism was constructed by support vector regression.A differential evolution algorithm was used to optimize the hyperparameters of the model.Results The data were changed from 729 variables to 20 variables after variable selection by random forest.The goodness of fit (R^(2)) and the mean square error (MSE) of the QSAR model based on support vector regression for drug antagonism ERα activity were 0.7407 and 0.7034.The QSAR model optimized by differential evolutionary algorithm had an R^(2) of 0.935 and MSE of 0.1762 for drug antagonistic ERα activity.Through the test of R^(2) and MSE,the optimized model was more robust and predicted accurately.Conclusion By establishing an QSAR model of compound molecular descriptors and their antagonistic ERα activity,we find the descriptors with important impact on the drug activity of breast cancer therapeutics,which can be used to guide the design of new breast cancer therapeutics.

关 键 词：乳腺癌 药物结构 定量构效关系 支持向量回归 差分进化算法 雌激素受体Α 

分 类 号：R96[医药卫生—药理学]