Research On Transmembrane Protein Prediction Based On SVM

Membrane protein is a sort of protein with important biological functions, and plays a vital role in organisms. Although phospholipids bilayer makes up of the basic framework in biological membrane, membrane protein is yet the main manifestation of biological membrane's function and it makes the material basis for cells to implement various functions. Predicting transmembrane protein whether it belongs to the classification ofβ-barrel transmembrane protein,α-helical transmembrane protein and inner membrane protein according to the sequence is the important precursor step for 3D structure modeling and function analysis and it is also a challenging task in computational protein field. In this study, we mainly explore the prediction of transmembrane classification.Feature extraction from transmembrane protein sequences is a basic problem for classification, and also is a key factor that determines the prediction performance. From the transmembrane proteins'primary sequences, we propose different feature extraction algorithms, and take some testing and analysis for these algorithms based on the selected dataset MCP1087 through 10-fold cross validation.We train three prediction models by support vector machine algorithm (SVM), and the features consist of the position information in sequence, the physical and chemical properties of amino acid residues. Three models could be used for the classification ofβ-barrel transmembrane protein,α-helical transmembrane protein and inner membrane protein. In the experiments, we explore more effective methods of feature extraction. At first, we calculate single amino acid index correlation coefficient, and then choose more optimal combination based on a single's correlation coefficient; Secondly, on the basis of the amino acid composition, the position sequence information of amino acid residues is added into features, to extract feature deeper from the protein's sequences, and to void the loss of amino acid's position information; At last, changing the order of the auto-correlation function can fully tap the long range correlation information of amino acid residues and further improve the prediction performance of the classification model.In this study, the prediction accuracy of three models in our testing sets separately can reach 88.36%, 87.89% and 94.41% in the best case. MCC related coefficient separately can reach 0.7723, 0.7667 and 0.8907. The experimental results presented show that transmembrane protein structure prediction based on SVM could provide valid enhancement to transmembrane protein 3D structure prediction and function analysis.