基于组合特征集成的蛋白质相互作用位点预测

运用RBF神经网络预测蛋白质相互作用位点.首先提取序列谱、保守权重、熵值、复合物可及表面积和序列变化率等一系列蛋白质相互作用位点的关键特征.然后应用RBF神经网络以及它们的集成来对这些样本集进行训练与测试.使用10次交叉验证进行训练与测试,创建了4组具有对比性的蛋白质相互作用特征组合.实验中每加入一种新的特征时正确预测率都会相应的提高,特别是加入可及表面积和序列变化率特征时正确率提高幅度更大,表明利用多特征组合,结合RBF神经网络算法进行预测蛋白质相互作用位点的方法是正确有效的.

Prediction of Protein-Protein Interaction Sites Based on Feature Combination Integration

This paper used the RBF neural network to predict protein-protein interaction sites.First,a series of features that can represent protein interaction sites,including sequence profiles,entropy,conservation weight,complex accessible surface area,sequence variability and so on,were extracted.Then,RBF neural network and their integrated approaches were applied to train and test these sample sets.In experiments,ten times cross-validation were used to predict the four sets of data respectively.The accuracy of prediction is proved when one new feature is added.When polar ASA and sequence variability input,the final result was more effective.The results show the validation and correctness of the method of combining feature amalgamation with RBF neural network.