Prediction Of Membrane Protein Types Based On Various Parameters From Sequence Information

Membrane protein could provide a functional place for biofilm, many membranes are considered as the targets of drugs and a majority of diseases are caused by the defect of certain specific membrane protein. The abnormity of membrane protein is closely related to cystic fibrosis, cancer, even senile dementia and Parkinson’s disease, etc.Undoubtedly, membrane protein function research plays a significant role in the development, design and screening of new drugs. The membrane protein functions are specifically corresponding to their categories and could be determined when the corresponding categories are identified. The category prediction is considered as a key means to determine the membrane protein functions. Membrane protein classification by laboratory experiment is featured by time-consuming, effort-consuming and high-cost,which is also limited by incapability of batch test. Therefore, it is of great necessity to develop a fast and reliable method to predict the membrane protein categories, which could provide a better service for human life. In this paper, amino acid sequence is taken into account to predict the membrane protein categories and the following researches are completed:(1) According to the membrane protein properties, the amino acid sequences of8-category membrane proteins are evenly divided and increment of diversity algorithm is used to predict the categories of 8-category membrane proteins based on the statistics of amino acid composition in each equal part.(2) The amino acid fragments in the N-terminals and C-terminals of 8-category membrane proteins are selected and scored by position matrix scoring to predict the categories of these8-category membrane proteins.(3) The combination values of increment of diversity, amino acid fragment scoring,rare motif frequency, self-intersection covariance, secondary structure frequency,super-secondary structure frequency and complex super-secondary structure frequency are taken as characteristic parameters and a support vector machine algorithm is used to predict 8-category of membrane proteins. Excellent prediction result is gained with an independent test and the overall prediction accuracy was 93.9%. The prediction accuracy of each category was as follows: Category-Ⅰ was 86.7%, Category-Ⅱ was 50.0%,Category-Ⅲ was 33.3%, Category-Ⅳ was 66.7%, Category-Ⅴ(multipass transmembrane) was 98.2%, Category-Ⅵ(lipid chain-anchored membrane) was 81.6%,Category-Ⅶ(GPI-anchored membrane) was 73.9% and Category-Ⅷ(peripheral membrane) was 81.5%.