基于监督学习算法的延胡索成分-靶点-疾病网络的预测研究

Study on prediction of compound-target-disease network of Corydalis yanhusuo based on supervised learning

目的应用监督学习算法中的随机森林算法构建药物-靶点模型,预测延胡索治疗心脑血管类疾病的关键靶标。方法训练集从KEGG数据库获取,包括药物-酶、药物-离子通道、药物-G蛋白、药物-核蛋白四类药靶数据,采用随机森林算法构建药靶模型并预测延胡索体内作用靶点,应用Cytoscape软件构建延胡索成分-靶点-疾病网络;其中模型精度利用十折交叉验证进行评价。结果四类药靶模型的预测正确率分别是71.36%、67.08%、73.71%、68.22%;延胡索的20个化学成分被预测出作用于ADRA2A、ADRA2C、ADRB2、ADRA1D、ADRB1、ADRA2B、DRD2、CACNA1B、GABRG2、SCN1A等多个与心脑血管疾病相关的靶点并得到较好的文献验证,每个化合物的平均靶点数为9.8。结论随机森林算法所建模型具有较好的预测正确率,能预测出延胡索治疗心脑血管类疾病的多个关键靶点,其还可用于预测其他中药化学成分的潜在作用靶点。

Objective To build the drug-target interaction model based on random forest algorithm of a supervised learning,and to predict key targets of Corydalis yanhusuo on cardio-cerebral vascular diseases.Methods The datasets of drug compounds and the related enzymes,ion channels,G-protein-coupled receptors,nuclear receptors downloaded from KEGG database were used as the training set.Random forest algorithm was applied to build the drug-target interaction model and to predict the potential targets interacted with the reported compounds of Corydalis yanhusuo.Cytoscape was used to construct a compound-target-disease network.The models accuracies were evaluated by 10-fold cross-validation tests.Results The overall success rates of the four models were 71.36%,67.08%,73.71%,and 68.22%,respectively.The predicted targets of 20 compounds of Corydalis yanhusuo on cardio-cerebral vascular diseases,including ADRA2A,ADRA2C,ADRB2,ADRA1D,ADRB1,ADRA2B,DRD2,CACNA1B,GABRG2,SCN1A,were validated by literatures.The average number of targets for each compound was 9.8.Conclusion The models established in this paper based on random forest algorithm have a good prediction accuracy,which can predict key targets of Corydalis yanhusuo on cardio-cerebral vascular diseases successfully.This method can be used to discover potential targets in other traditional Chinese medicine ingredients.