基于BP神经网络的抗乳腺癌候选药物预测

Prediction of Anti-Breast Cancer Drug Candidate Based on BP Neural Network

摘要: 目前，乳腺癌已成为世界上致死率最高的癌症之一。据研究表明，通过调节雌激素受体亚型(ERα)活性可以控制人体内的雌激素水平，人体内的雌激素水平与乳腺癌的发展密切相关。而利用可以拮抗ERα活性的化合物有极大的可能可以治疗乳腺癌疾病，比如，临床治疗乳腺癌的经典药物他莫昔芬和雷诺昔芬就是ERα拮抗剂。本文基于化合物对ERα活性和化合物分子描述等已有数据，建立一个用于预测化合物对ERα生物活性的定量预测模型，然后利用随机森林和皮尔逊相关系数对影响生物活性最大的变量进行筛选降维，利用降低维数后的变量，搭建BP神经网络，构建预测模型。最后通过测试得到我们的模型预测效果好，泛化能力强，不容易发生过拟合。并对所提出的预测模型进行展望与分析。

Abstract: Breast cancer has become one of the world’s most deadly cancers. Studies have shown that estrogen levels in the human body, which are closely associated with the development of breast cancer, can be controlled by regulating the activity of the estrogen receptor sub-type (ERα). Therefore, the use of compounds that antagonize ERα activity has great potential to treat breast cancer diseases. For example, tamoxifen and raloxifene, the classic drugs in clinical treatment of breast cancer, are ERα antagonists. Based on compounds of ER alpha activity and molecular description of existing data, set up a used to predict compounds quantitative prediction model of ER alpha biological activity, and then use random forests and Pearson correlation coefficient of the biggest variables to affect the biological activity screening dimension reduction, after using the lower dimension variable and building a BP neural network, forecast model was constructed. Finally, through testing, our model has good prediction effect, strong generalization ability is not prone to overfitting. The prediction model is prospected and analyzed.