基于BP神经网络的ERα生物活性定量预测模型

Quantitative Prediction Model of ERα Bioactivity Based on BP Neural Network

摘要: 本文对抗乳腺癌候选药物优化问题进行了QSAR建模，使用MATLAB软件进行基于BP神经网络的ERα生物活性定量预测模型构建及预测。构建三层BP神经网络，经过莱文贝格–马夸特法和量化共轭梯度法分别模拟10次，选择最终模型。该模型显示，通过17次迭代后残差收敛，R²均达到0.75以上，均方误差为0.787，模型拟合较好。随后使用模型对test表中50个化合物进行生物活性预测，并将结果填入对应列。

Abstract: In this paper, the QSAR model was established for the optimization of candidate drugs for breast cancer, and the quantitative prediction model of ERα biological activity based on BP neural network was constructed and predicted by MATLAB software. The three-layer BP neural network is constructed, and the final model is selected after 10 times simulations by the Levinberg-Maquart method and the quantitative conjugate gradient method. The model shows that after 17 iterations, the residuals converge, R² reaches more than 0.75, and the mean square error is 0.787. The model fits well. Subsequently, the model was used to predict the biological activities of 50 compounds in the test table, and the results were filled in the corresponding column.