中药化学成分神经毒性预测定量构效关系模型构建OA

Objective To construct a quantitative structure-activity relationship(QSAR)model for predicting the potential neurotoxicity of traditional Chinese medicine chemical components.Methods A total of 1 769 compounds with neurotoxicity and 596 compounds without neurotoxicity were collected from the Toxicity Reference Database(ToxRefDB),Side Effect Resource(SIDER)databases and literature,to constitute the training set for QSAR model construction.The RDkit 2022.09.5 software package was used to calculate and screen molecular descriptors.Five algorithms including K-Nearest Neighbors,Naive Bayes,Random Forest,XGBoos and Artificial Neural Network,were employed to construct QSAR models,and internal validation was performed using a 10 fold cross validation method to select the optimal model.11 compounds with neurotoxicity and 13 compounds without neurotoxicity were collected as external validation sets by consulting literature and databases.This set was used to evaluate the applicability of the developed QSAR model for predicting traditional Chinese medicine chemical components neurotoxicity.Results After internal and external verification,Random Forest and ANN performed outstandingly in the balanced recognition of positive and negative examples among the five algorithms.Among all drugs,the ANN model correctly predicted 16 instances and incorrectly predicted 8.The Random Forest model accurately predicted 17 and incorrectly predicted 7,with both models achieved a prediction accuracy of over 65%.Conclusion The predictive ability of models constructed using Random Forest or ANN algorithms is superior to those constructed using K-Nearest Neighbors,Naive Bayes,and XGBoos algorithms.Validation of known toxic traditional Chinese medicine chemical components shows that this QSAR model has good sensitivity and prediction accuracy,indicating its applicability for predicting the neurotoxicity of traditional Chinese medicine chemical components.