基于序列剖面和可及表面积的蛋白质相互作用位点的预测

Predicting Protein-Protein Interaction Sites from Sequence Profile and Accessible Surface Area

蛋白质相互作用位点的预测对于突变设计和蛋白质相互作用网络的重构都是至关重要的.由于实验确定的蛋白质复合物和蛋白质配体复合物的结构依然相当少,预测蛋白质相互作用位点的计算方法就显得十分重要.该文提出了一种以支持向量机为分类器,以邻近残基的序列剖面和可及表面积为输入数据来预测蛋白质相互作用位点的方法.计算结果显示,界面残基和非界面残基被识别的准确率为75.12%,假阳性率为28.04%.与输入数据仅有序列剖面的方法相比,界面残基和非界面残基被识别的准确率提高了4.34%,假阳性率降低了4.63%.

Prediction of protein-protein interaction sites is essential for mutant design and reconstruction of protein-protein interaction networks. Because the number of experimentally determined structures for proteinprotein and protein ligand complexes is still quite small, methods for computational prediction of proteinprotein interaction sites are becoming increasingly important. Protein-proteln interaction sites are predicted using support vector machines （SVM） with sequence profiles of neighboring residues and accessible surface area as input. Interface residues and non-interface residues were identified with relative accuracy of 75.12% and a false positive rate of 28.04%. The accuracy is 4.34% higher, and the false positive rate 4.63% lower, than that obtained with only sequence profile for the feature vectors for the SVM.