Study Of G-protein Coupling Specificity Prediction Based On Support Vector Machines

G protein coupled receptors (GPCRs) comprise a superfamily of tranmembrane proteins in eukaryotes. GPCRs play an important role in signal transduction from the extracellular space into the innercell across the cell membrane via coupling to G protein. GPCRs also are a major drug target for the pharmaceutical industry because more than 50% of current prescription drugs are act on GPCRs. At present, many drugs for GPCRs only act on 30 kinds of GPCRs, which are a small part of the GPCRs family. Therefore, a large majority of GPCRs still remain promising drug targets. Research on the coupling specificity of GPCRs-G protein is critical for further understanding the function of GPCRs and the mechanism of cellular signal transduction, which can provide new clues for pharmaceutical research and development.In order to maximize mining the coupling information in GPCRs sequences, this research gain coupling information of samples from gpDB database, then varieties of features were extracted from different domains of GPCRs sequences, classifiers for non-promiscuous coupling and promiscuous coupling were developed respectively using support vector machines. The main innovations of this study are list as follows.(1) All of the coupling information of GPCRs to G proteins has been collected from gpDB database as our dataset, which contined more GPCRs sequence than other published methods.(2) The impact of every innercell domains of GPCRs to coupling specificity is different. The experiments were conducted on each region respectively. There is not related content in previous studies.(3) Two classifiers for non-promiscuous coupling and promiscuous coupling have been developed respectively. Almost of the previous research are restricted to non-promiscuous coupling.In the experiment of non-promiscuous coupling specificity,10-fold cross validation and independent dataset were used to evlaue the performace of the model, which yield a total accuracy of 91.8977% and a total accuracy of 96.063% respectively. In the prediction of promiscuous coupling specificity,5-fold cross validation and Jackknife test were used to evlaue the the performace of the model, which achieved a total accuracy of 84.4595% and a total accuracy of 86.4865% respectively. The results show that the method of our study is effective in the prediction of G protein coupling specificity.