Predicting Potential Drug Targets Of G-protein Coupled Receptors Based On SVM

Object: The key to the success of drug development is the position ofdrug target. However the number of clinical approved drug targets is small. Itis is urgent to find more drug targets. G protein-coupled receptors are thevast majority of the known drug targets. They are closely linked to manydiseases, such as high blood pressure, asthma, pain,neurological andimmune disorders. Because GPCRs’ seven transmembrane conformation iscomplex, the spatial structure of GPCRS are difficult to obtain from theexperiment. Therefore, their functions are difficult to determine. In this study,we combined protein sequences,peptides and protein basic physical andchemical properties and other characteristics to construct support vectormachine (SVM) classifiers to predict the potential drug targets in GPCRs.This study can provide theoretical support for drug development.Method: In this study, we use two different classes protein sequences tobuild SVM classifiers. These protein sequences are download from threedatabase, GPCRDB, Uniprot and DrugBank. The first classifier isrecognized by the human drug targets as positive set, filtered non-drugtargets as negative set. The other type of classifier is recognized GPCR drug targets as positive set and filtered GPCR non drug targets as negative set. Weextracted the primary sequence characteristic, polypeptide characteristic andphysical and chemical properties characteristic of the protein as the featurespace of the training classifier. We use genetic algorithm to select thesecharacteristics. Though regulate the parameters of the model to build twosets of optimal classifiers. Finally, these two sets of optimal classifiers wereused to prediction the potential drug targets in GPCRs.Result: The classification accuracy rate is of the first class of classifiers72.63%, sensitivity is77.44%,and specificity is67.55%. The secondclassification accuracy, sensitivity, specificity are about95%.141GPCRsare predicted as drug target by the two classifiers. This proportion accountedfor17.5%of the entire GPCRs,.this results of GPCRs research have veryimportant significance.Conclusion: Building two sets of classifiers to predict GPCR drugtargets. The two sets of classifiers mutual authentication to increase thecredibility of the classification results.39proteins which predicted as drugtargets by the two sets of classifiers are also found in TTD(TherapeuticTarget Database). The result proved the feasibility and correctness of thismethod.