基于异构网络特征与梯度提升决策树的协同药物预测

组合药物在复杂疾病特别是癌症的治疗中发挥越来越重要的作用。以组合药物靶标为初始节点在药物-蛋白质异构网络上执行重启型随机游走,将收敛后的概率分布作为药物组合的特征向量,训练梯度提升决策树模型来预测新的药物组合。在标准药物组合数据集的性能评估表明,该方法比其他七种典型分类器和传统的提升算法具有更好的性能,且基于异构网络的特征显著提升了各分类器的性能,AUC值从0.528提升至0.909。

SYNERGETIC DRUG PREDICTION BASED ON THE FEATURES OF HETEROGENEOUS NETWORK AND GRADIENT BOOSTING DECISION TREE

Drug combination plays an increasingly important role in the treatment of complex diseases,especially cancer.In this paper,we took the target of drug combination as the initial node to perform restart random walk on the drug-protein heterogeneous network.Taking the converged probability distribution as the feature vector of drug combination,the gradient boosting decision tree model was trained to predict the new drug combination.The performance evaluation on benchmark drug combination dataset shows that our method achieves higher performance than the other 7 typical classifiers and traditional boosting algorithms.Furthermore,the features extracted from heterogeneous network significantly improve the performance of each classifier,especially the AUC value of our algorithm is increased from 0.528 to 0.909.