The research of algorithm for protein subcellular localization prediction based on SVM-RFE

In view of it is not only time-consuming but also costly to determine the locations of the protein subcellular by biological experiment, developing fast and effective calculation methods for subcellular localization prediction has become one of the important research contents in the field of bioinformatics. Since the SVM-RFE algorithm can select the optimal feature subset according to the correlation between each feature and protein subcellular localization, and it can reduce the computational complexity while keeping the result steady and having a high degree of generalization in the progress of using the RFE part of SVM-RFE method, therefore, this algorithm is applied to predict protein subcellular localization. First of all, we extract amino acid components, dipeptide component and entropy density from Position Specific Scoring Matrix to construct the feature expression model of protein sequence. Then we use the recursion feature elimination to conduct feature selection. Finally, the support vector machine classifier was used to conduct Jackknife verification on two data sets of Gram Positive and Negative. The experimental results show that the application of SVM-RFE algorithm to protein subcellular localization has a good predictive accuracy.