Drug-protein Interaction Prediction based on Improved AdaBoost Algorithm

Abstract

The drug-protein interaction prediction can be used in the discovery of new drug effects. Recent studies often focus on the prediction with an independent matrix filling algorithm. The single-model matrix-filling algorithm has low accuracy, so it is difficult to obtain satisfactory results in the prediction of drug-protein research. The AdaBoost algorithm is an algorithm framework of a powerful classifier composed of multiple subclassifiers. Its usefulness and effectiveness has been proved in the research field of classification. The prediction of drug-protein is a matrix filling problem, which is a process of scoring prediction. Therefore, we improved the algorithm of AdaBoost. To transform the matrix filling problem into a classification problem with improved AdaBoost algorithm. The AdaBoost algorithm framework could be fully utilized to integrate multiple weak classifiers to improve prediction performance. Then we can make an accurate prediction of the drug-protein interaction. The experimental results based on the public data set show that the proposed algorithm outperforms most classical and recent algorithms in predicting accuracy. The limitation of single algorithm based on machine learning is overcome well. Our method improves the accuracy of prediction by mining the hidden factors better.