基于融合核方程对药物-靶点作用预测研究

Prediction of drug-target interactions by integrated kernel function

利用计算方法预测药物-靶点作用能够有效降低药物研发成本。本文提出一种基于支持向量机的新算法预测药物-靶点信息。该方法通过整合药物-靶点作用信息、药物分子物化特性、蛋白质序列信息,对药物-靶点作用对进行特征提取,并设计新的融合核方程,将药物分子、蛋白质分子信息映射到同一个空间,利用支持向量机对药物-靶点作用进行预测分类。10-CV交叉验证下,本文方法预测总精度为93.25%,与Van-Laarhoven高斯作用核方法相比提高了6.65%。实验结果表明,本文方法可以有效预测药物-靶点之间潜在作用。

Prediction of drug-target interactions by computational methods is an important part of drug discovery. In this paper, we propose a novel method based on support vector machine to predict drug-target interactions by integrating drug-target interaction information, biochemical properties of drugs and protein sequence similarity. By employing the integrated kernel function, drugs and targets are mapped into a unified space and predicted by support vector machine classifier. The prediction accuracy of the proposed method is 93.25%, which is 6.65% higher than that of Van-Laarhoven＇s Gaussian interaction profile kernels with 10-CV test. The results show that the proposed method is effective to infer the potential drug-target interactions.