Prediction Of Functional Sites And SNPs In G Protein-coupled Receptors

G protein-coupled receptors(GPCRs),which also are called seven a-helices transmembrane(TM) receptors,are one of the largest superfamily of proteins. GPCRs contain seven TM regions separated by six loops:three extracellular (ECL) and three intracellular(ICL),with an amino terminus outside the cell and a carboxyl terminus inside the cell.The completion of the human genome sequencing projects has identified about 1200 genes that belong to the GPCR superfamily.These receptors respond to a variety of ligands as diverse as photons, bioamines,peptides,glycoproteins,lipids,nucleotides and even large proteins, which felt by the heterotrimeric G protein at the cytosolic side.This G protein in turn modulates the activity of enzymes by dissociation of its a andβγsubunit complex,activates a second messenger system,transmits signals from outside to the effectors inside and makes changes in the cell.The malfunction of GPCRs could lead to many severe diseases,for example,the Alzheimer disease, Parkinson disease,depression,hypertension,diabetes and schizophrenia.So GPCRs play an important role in pharmaceutical realm.At present they are targets for about 1/3 of all known small molecule drugs in world market.In this dissertation,we aimed to investigate the functional sites and functional nsSNPs in GPCRs by in silico method since they have an important meaning in drug design.All of the GPCRs have similar structures and share a common mechanism of activation,but they mediate different physiological functions because of different residues playing different roles in GPCRs.In this dissertation,we investigated first the bioaminergic receptors,one subfamily of GPCRs,and built the predictive model for functional sites.According to subfamily definition,the model identified the global and subfamily-specific functional sites based on the position-specific conservation score in both family and subfamily.Then the functions of predicted functional sites were analyzed based on the existed mutagenesis experiments. Finally,the prediction model for functional sites was applied to the GPCRs superfamily.The following results were derived:(1) Global functional sites which were mainly located in the TM regions were highly conserved in the bioaminergic receptors and played key roles in structure and responsible for the common mechanisms of receptor activation and G protein coupling.(2) Residues conserved in one subfamily but varied between other subfamilies were closely related with the specific functions of subfamily.Subfamily-specific residues were different at the quantity and distribution.They mainly located in TM regions,the second ECL and ICL and involved in ligands binding and G protein coupling.The functional nsSNPs in GPCRs were predicted by machine learning methods.The predictive power of each attribute was assessed by 1R algorithm, X~2-statistics and information gain,respectively.The attribute subsets were forward searched by genetic algorithm based on correlation,consistency and wrapper algorithm,respectively.The prediction performance of decision tree, support vector machine and K nearest neighborhoods was evaluated on different attributes sets,then the optimal attribute set and classification method were selected.The following results were derived:(1) From the single attribute aspect, the conservation score of the mutated position was the best predictor to distinguish functional mutations from neutral ones,while the predicted structural attributes did much worse.Combining the sequence and structural attributes can improve the classification performance,but simply taking all attributes together would not achieve the best.(2) The wrapper-based attribute subset was the optimal attribute set for prediction functional nsSNPs.It was consisted by six attributes,which were conservation score of the mutated position,the hydrophobicity change between wild-type residue and mutated residue, BLOSUM62 substitution matrix score,the location of the mutated position, relative solvent accessibility and buried charge.Therefore,for the proteins with unknown structures,predicted structural information was also useful in classification.(3) Decision tree method combined with optimal attribute set achieved the prediction accuracy 91.17%,and the model was generalization.The decision tree method can not only achieve the highest prediction performance among three methods,but also produce intelligible rules with the prediction accuracy attached to each rule.In this dissertation,30 rules were derived from the decision tree,among which 16 rules were used to predict the functional SNPs and 14 rules were used to predict the neutral SNPs.(4) A total of 519 non-synonymous SNPs in human GPCRs were collected from dbSNP,and 166 SNPs have been predicted to be functional with the optimal attribute set by decision tree method.Further analysis of these SNPs will provide a basis for assessing susceptibility to diseases and designing individualized therapy.