基于流形正则化非负矩阵分解预测药物-靶蛋白作用关系

识别药物-靶蛋白作用关系是当前药物研究的重要内容,其可帮助识别已有药物的新功能,发现药物的"偏靶蛋白"等。现有预测算法对新药物的作用靶蛋白,及新靶蛋白的作用药物预测存在困难,由此提出一种新奇的基于流形正则化非负矩阵分解的新药物/新靶蛋白作用关系预测算法,该方法首先通过聚类算法构建新药物/新靶蛋白的初始作用标签,然后设计引入流形学习正则化约束的非负矩阵分解算法预测药物-靶蛋白作用关系,最后在四个经典数据集中测试,并与最新预测算法BLM-NII、RLS-WNN和WKNKN+WGRMF算法进行比较,证明本文算法可获取较高的预测精度。

Prediction of Drug-Target Interaction with Manifold Regularized Non-Negative Matrix Factorization

Identification of drug-target Interactions (DTIs) is very important for drug research, which can help to find the new uses for old drugs or to discover the off-target of a given drug. Currently, the prediction algorithms have difficulty in finding interactions for new drugs and new targets. We proposed a novel method that uses manifold regularized nonnegative matrix factorization framework to predict potential targets/drugs for new drugs/targets. Firstly, it used clustering approaches to construct interaction profiles for new drugs/targets; then adopted the manifold regularized nonnegative matrix factorization algorithm to predict the drug-target Interaction; Finally, extensively testing was applied on four datasets. Through comparison with other recently proposed BLM-NII,RLS-WNN and WKNKN+WGRMF, our algorithm attains high prediction performance in terms of AUPR.