基于BP神经网络的抗乳腺癌药物的选择与寻优
Selection and Optimization of Anti-Breast Cancer Drugs Based on BP Neural Network

研究表明，雌激素受体α亚型(Estrogen receptors alpha, ERα)在乳腺发育过程中至关重要，ERα被认为是治疗乳腺癌的重要靶标。本文基于1974个与ERα的生物活性有关的化合物，采用机器学习，构建化合物生物活性的定量预测模型和ADMET性质的分类预测模型。随后选用基于样本相关系数的检验对729个分子描述符对生物活性影响的重要性进行排序，最终得到了前20个最具影响的分子描述符。采用具有非线性映射能力的BP神经网络来建立生物活性预测模型。同时从线性模型与非线性模型两个角度出发来构建模型，计算得出两种模型寻找的20个主要分子描述符及获得的生物活性和ADMET性质。
Studies have shown that estrogen receptors alpha (ERα) is crucial in mammary gland development, and ERα is considered an important target for breast cancer treatment. Based on 1974 compounds related to the biological activity of ERα, this paper uses machine learning to build a quantitative prediction model of compound biological activity and a classification prediction model of ADMET properties. Then, the test based on the sample correlation coefficient was used to rank the importance of the influence of 729 molecular descriptors on biological activity, and finally the top 20 most influential molecular descriptors were obtained. A BP neural network with nonlinear mapping ability was used to establish a biological activity prediction model, and a multiple linear regression model and a gradient boosting regression tree model were established for comparison and verification. At the same time, the model was constructed from the perspective of linear model and nonlinear model, and the 20 main molecular descriptors sought by the two models and the obtained biological activities and ADMET properties were calculated.