Research Of Transmembrane Protein Topology Prediction Based On Multi-source Information Fusion

According to genome-wide estimations, roughly20-30%of the genes in a typical organism code for transmembrane (TM) proteins. Transmembrane proteins are some special and important kinds of proteins, which are the principal executives of the function of biomembrane, and play many important and functionally diverse roles in cells, such as energy conversion, substance transportation and information transfer. The study of transmembrane protein topology is very significant to discover transmembrane mechanism and function. Due to the limitation of current technical level and conditions, experimental method to determinate the topology structure of transmembrane protein cannot keep up with the rapid increase of transmembrane sequences information. Therefore, various computational methods have been developed to predict the topology of transmembrane protein, and the topology prediction of transmembrane protein has become a hot field in bioinformatics.At present, most prediction methods of transmembrane protein topology are on the basis of the chemical or physical properties of amino acids such as the hydrophobicity of residues and "positive-inside rule", or on the basis of some theoretical methodolgoy such as statistic analysis and machine learning. However, these methods mostly are limited to the use of one or several distinguishing features of transmembrane protein, and they do not make an all-round and comprehensive use of various potential useful information. The prediction ability still needs to be improved. Moreover, the study to protein’s high-level structure requires more accurate and effective prediction methods. Therefore, based on the idea of multi-source information fusion, an integrated combination prediction platform of transmembrane protein topology has been established by applying evidence theory to the transmembrane protein topology prediction. The effectiveness of proposed method has been demonstrated through three levels, namely amino acid residue prediction, transmembrane regions prediction and topology prediction, respectively. The proposed method has provided new research thoughts and a powerful tool for the study of transmembrane protein structure.In addition, on the process of establishing the evidence theory based transmembrane protein combination prediction platform, the construction of basic probability assignment (BPA) is the capital problem, which is also the most important and hardest in the application of evidence theory. In order to solve this problem, a new confusion matrix based method is proposed to construct the BPA on the basis of summarizing the previous studies. The proposed BPA construction method makes full use of the available information contained in the confusion matrix, and fully reflects the recognition ability of classifier to each class. The proposed method is also very significance for the reference of evidence theory.