抗乳腺癌候选药物的优化建模

Optimal Modeling of Anti-breast Cancer Drug Candidate

摘要: 雌激素受体α亚型(ERα)是临床重要的药物靶点，治疗乳腺癌候选药物的化合物需能拮抗ERα活性。采用建立化合物活性预测模型的方法来筛选乳腺癌候选药物可降低药物研发的时间和成本。本文先利用随机森林对分子描述符对生物活性的贡献度排序，然后按变量相关性进行系统聚类，采用斯皮尔曼相关系数确定与生物活性相关性最显著的20个变量。接着利用神经网络建立化合物对ERα生物活性的定量预测模型，对比预测与实际结果。然后利用二次SVM构建化合物ADMET性质分类预测模型。最后，利用粒子群算法进行目标寻优，确定分子描述符及其取值使得生物活性和ADMET性质最优。本文模型能很好预测具有更好生物活性的新化合物分子，能实现可作为临床治疗乳腺癌候选药物的化合物的筛选。

Abstract: Estrogen receptors alpha (ERα) is an important clinical drug target, and the candidate drug com-pounds for breast cancer should be able to antagonize ERα activity. The method of establishing a compound activity prediction model to screen candidate drugs for breast cancer can reduce the time and cost of drug development. In this paper, the contribution degree of molecular descriptors to biological activity is ranked by random forest, and then systematic clustering is carried out ac-cording to the correlation of variables. Spearman correlation coefficient is used to determine the 20 variables with the most significant correlation with biological activity. Then, establishing a quantitative prediction model for the biological activity of ERα by neural network, and comparing the prediction with the actual results. Then the ADMET property classification prediction models are constructed by quadratic SVM. Finally, particle swarm optimization (PSO) algorithm is used to optimize the target, and the molecular descriptors and their values are determined to optimize the biological activity and ADMET properties. The proposed model can well predict novel compound molecules with better biological activity and realize the screening of compounds that can be used as candidate drugs for clinical treatment of breast cancer.