Prediction of Potential Targets of Traditional Chinese Medicine Based on Machine Learning

Purpose: taking Lateralis Radix Praeparata which is in common use for example, explore to predict the target with Chinese medicine composition and construct Chinese medicine multi-component - multi-target network. Method: This study collected 2388 medicine molecule structure and target data which was released by America FDA for sale in drugbank database, coding the data by use of PowerMV and k-spaced, establishment of the interacted medicine and target model based on SVM. Evaluation of the predicted model performance by use of five folder cross method, the average accuracy of the model for training data set can achieve 79.74% and can be 82.41% for independent training data set. Prediction of target for the Lateralis Radix Praeparata composition by use of the model. Results: Prediction of several targets by use of the 24 compositions of Lateralis Radix Praeparata. The average target number of each compound in the network model is 63.42, each target linked with 7.42 compounds which embodied the Chinese medicine big feature for multi-composition and multi-target. Conclusion: This method can be used to identify some potential targets of traditional Chinese medicine.