基于PSO-BP的抗乳腺癌药物毒性研究

Study on Toxicity of Anti-Breast Cancer Drugs Based on PSO-BP

为解决新药研发过程中药物的毒性难以准确预估的问题,利用计算机技术,提出一种基于粒子群算法(PSO)优化BP神经网络的二分类预测模型。通过互信息的方法从729个分子描述符中筛选出重要度最高的20特征作为自变量,以药物的毒性值作为因变量,在BP神经网络模型的基础上,首先使用不同的梯度下降算法计算模型的准确率,发现批量梯度下降算法对BP模型的拟合效果较好;其次利用动态变权重的粒子群算法对BP神经网络模型的权重和阈值进行优化选择,结合BP神经网络、SVM和KNN模型进行对比实验,结果显示,PSO-BP模型的准确率、精确率、召回率和F1值明显高于其它模型。因此,PSO-BP模型是一种对抗乳腺癌药物毒性有效预测的方法。

In order to solve the problem that it is difficult to accurately predict the toxicity of drugs in the process of new drug research and development,a binary prediction model based on particle swarm optimization(PSO)optimized BP neural network is proposed by using computer technology.In this paper,the 20 features with the highest importance were selected from 729 molecular descriptors by mutual information method as independent variables and the toxicity value of drugs as dependent variables.Based on the BP neural network model,firstly,different gradient descent algorithms were used to calculate the accuracy of the model.It was found that the batch gradient descent algorithm has the best fitting effect on the BP model.Secondly,the weights and thresholds of the BP neural network model were optimally selected by using the particle swarm algorithm with dynamically variable weights,and the comparative experiments were carried out with the BP neural network,SVM and KNN model,and the results showed that the accuracy,precision,recall and F1 value of the PSO-BP model were significantly higher than that of other models.Therefore,the PSO-BP model is an effective method to predict the toxicity of anti-breast cancer drugs.