基于理化性质局部并行融合的蛋白质相互作用预测方法

Prediction of Protein Interactions Based on Local Parallel Fusion of Physicochemical Properties

针对从蛋白质的原始序列中提取特征向量的预测精度不高,提出了基于理化性质局部并行融合的特征提取方法。首先,从蛋白质的原始序列计算PSSM矩阵,根据理化性质将PSSM矩阵分成4个区域,进行并行融合;然后,构建蛋白质序列的特征向量,采用主成分分析法提取关键特征信息,构建蛋白质序列对的特征向量。建立了多项式核和高斯核组合核支持向量机预测模型,使用酿酒酵母数据集进行测试,该方法达到93.98%的预测准确率,预测效果优于原始序列特征提取方法。使用独立数据集进行模型泛化性验证实验,也同样表现出了良好的性能。

Due to the fact that the low prediction accuracy of extracting feature vectors from the original sequence of proteins,the present study proposed a feature extraction method based on local parallel fusion of physical and chemical properties.First,the PSSM matrix was calculated according to the original protein sequence,and the PSSM matrix was divided into four regions according to the physical and chemical properties for parallel fusion.Then the feature vectors of each protein sequence are constructed.Principal component analysis was used to extract key feature information and construct feature vectors of protein sequence pairs.The prediction model of polynomial kernel and Gaussian kernel combined kernel support vector machine is established.Experimental results show that the prediction accuracy of this method is 93.98%on Saccharomyces cerevisiae dataset,which is better than the original sequence feature extraction method.It also shows good performance when using independent data sets to verify the model generalization.