基于随机森林回归的乳腺癌抗药生物活性预测

Prediction of Antibiotic Activity of Breast Cancer Drug Resistance Based on Random Forest Regression

摘要: 在药物研发中，雌激素受体α亚型(Estrogen receptors alpha, ERα)被认为是治疗乳腺癌的重要靶标，能拮抗ERα活性的化合物可能是治疗乳腺癌的候选药物。因此本文旨在以生物活性值pIC50作为因变量，作用于ERα靶标化合物的分子结构描述符作为自变量，构建关于ERα靶标化合物的生物活性预测模型，进而挑选出有效的抗癌候选药物。首先采用方差过滤法、随机森林、XGBoost以及灰色关联分析对自变量进行筛选，得到MDEC-23等16个与pIC50相关性强，且变量间相关性弱的分子结构描述符。其次建立随机森林回归生物活性预测模型，将预测结果与支持向量回归、梯度提升回归树、XGBoost模型和MLP回归模型预测结果进行对比分析，结果表明随机森林回归模型能更好地拟合数据，在R2、MAE、MSE上优于其它模型，更适应于对生物活性pIC50值的预测，同时也表明筛选出的分子结构描述符在一定程度上能治疗乳腺癌。

Abstract: In drug development, the Estrogen receptors α subtype (ERα) is considered to be an important target for the treatment of breast cancer, and compounds that antagonize ERα activity may be candidates for the treatment of breast cancer. Therefore, this paper aims to use the biological activity value pIC50 as the dependent variable and the molecular structure descriptor acting on the ERα target compound as the independent variable to construct a prediction model for the biological activity of the ERα target compound, and then select effective anti-cancer drug candidates. Firstly, the independent variables were screened by variance filtering, random forest, XGBoost and gray correlation analysis, and 16 molecular structure descriptors such as MDEC-23 with strong correlation with pIC50 and weak correlation between variables were obtained. Secondly, a random forest regression biological activity prediction model is established, and the prediction results are compared and analyzed with the prediction results of support vector regression, gradient boosting regression tree, XGBoost model and MLP regression model, and the results show that the random forest regression model can better fit the data, is better than other models in R2, MAE and MSE, and is more suitable for predicting the bioactive pIC50 value, and also shows that the screened molecular structure descriptors can treat breast cancer to a certain extent.