Prediction Of Protein Solvent Accessibility Based On All-atom Encoding

While explain the structure of proteins and biological function, the surface characteristics of protein is an important tool. In folding process, the hydrophobic group of protein tend to buried in molecular internal, the study of proteins solvent accessible surface can get more information of protein folding and hydrophobic. Because solvent accessible surface maybe big or small, simply through its value can't accurate measure its physical properties, so we change solvent accessible surface into solvent accessibility, further to analyze the structure and property of protein. According to theoretical modeling predict protein solvent accessibility, can reduce the costs of experiment and improve the calculation efficiency of protein solvent accessibility. In order to further research protein solvent accessibility which based on sequence, this paper explores a new kind of coding way-all-atom coding.Based on support vector machine method, this paper used three different encoding schemes to predict protein solvent accessibility, they are amino acid sequence encoding, amino acid descriptor encoding and all-atom encoding respectively. For common data set RS126, based on amino acid sequence encoding model, mean absolute error and correlation coefficients is 18.9%,0.527 respectively:Based on amino acids descriptor encoding model, mean absolute error and correlation coefficient is 19.2%,0.508 respectively; And all-atom encoding model, mean absolute error and correlation coefficients is 18.7%,0.528 respectively. The results demonstrate all-atom encoding is superior to other encoding methods, it has some extensibility.The innovation of this paper is put forward all-atom encoding, according to statistics, the atomic structured of 20 amino acids summarized by 215 kinds of atoms, to encoding 215 kinds of atoms by binary, get all-atom encoding method, and applied it in the prediction of protein solvent accessibility. It can be seen that the information of protein sequence which obtained by experiments is more than the information of protein structure, so encoded known finite protein structure, obtain the dependency of structural and solvent accessibility, and then establish the calculation model of unknown structure protein residues solvent accessibility. The model is constructed by all-atom encoding, which is more intuitive and easier to explain the contribution of every atom to solvent accessibility.