【摘要】 为了提高G蛋白偶联受体(G-protein-coupled receptors,GPCR)与药物相互作用预测的精度,该文提出一种基于多视角特征组合与随机森林的GPCR-Drug相互作用预测新方法。该方法首先从氨基酸组成成分和蛋白质进化视角分别抽取GPCR的序列特征,并从分子指纹视角抽取药物分子的特征;将所抽取的多视角特征进行组合,得到GPCR-Drug配对的特征表示;基于所提出的GPCR-Drug特征表示方法,使用随机森林构建预测模型。在标准数据集上的交叉验证和独立测试结果验证了该文所述方法的有效性。更多还原

【Abstract】 In order to improve the accuracy of predicting the interactions between G-protein-coupled receptors( GPCR) and drugs,this paper develops a novel method based on multi-view feature combination and random forest for GPCR-Drug interactions prediction with high performance. In the method,GPCR features from amino acid composition and protein evolution views and drug feature from molecular fingerprint are extracted; the feature of every GPCR-Drug pair can be formulated by serially combining the multi-view features of GPCRs and drugs; the GPCR-Drug prediction model is constructed with the random forest algorithm under the developed feature representation. Stringent experiments on benchmark datasets over both cross-validation and independent validation tests demonstrate the feasibility and efficacy of the proposed method.更多还原