论文部分内容阅读
该文通过计算324个神经毒性化合物和235个无神经毒性化合物的物理化学性质、电荷分布及几何结构等特征的6 122个分子描述符,通过CfsSubsetEval评价和BestFirst-D1-N5搜索相结合的方法筛选描述符,利用支持向量机(SVM)构建了化合物神经毒性判别模型。模型的准确率、灵敏度、特异性均在80%以上。以30个确有神经毒性的中药成分作为外部验证集,进一步验证模型准确率,达73.333%。将该模型应用于山豆根神经毒性成分筛查,筛得13个潜在神经毒性化合物,其中4个已有文献验证。实验结果表明该模型具有一定的准确性,有助于开展中药神经毒性成分筛查工作。
In this paper, 6122 molecular descriptors of physicochemical properties, charge distribution and geometric structure of 324 neurotoxic compounds and 235 non-neurotoxic compounds were calculated by combining CfsSubsetEval evaluation and BestFirst-D1-N5 search The descriptors were screened and the compound neurotoxicity discriminant model was constructed by using Support Vector Machine (SVM). The accuracy, sensitivity and specificity of the model are above 80%. 30 authentic TCM ingredients with neurotoxicity were used as external validation sets to further verify the accuracy of the model, reaching 73.333%. Thirteen potential neurotoxic compounds were screened for the application of this model in the detection of neurotoxicity components of Radix solanum, four of which have been documented. The experimental results show that the model has a certain accuracy, which is helpful for the screening of neurotoxic components of traditional Chinese medicine.