Study Of G-protein Coupled Receptor Prediction Based On Hidden Markov Model

G-protein coupled receptor(GPCR), composed of a polypeptide chain thrustinginto the cell membrane up to seven times, belongs to one of the largest superfamiliesof membrane protein in human body. As a common drug target molecule, itaccomplishes cellular signal transduction by combining it with signal molecules toactivate G protein. The dysfunction of GPCR relates to various of diseases. Thus,further study of GPCR is significant for understanding signal transduction mechanismas well as new drug development. The number of known sequence of GPCR iscurrently much less than that can be encoded by human genome. So there is a lot ofwork in identifying new GPCR for candidate drug target. However, traditionalexperimental methods for determining GPCR are much expensive and inefficient.Therefore, predicting GPCR families with bioinformatic methods shows greatresearch and application value.Based on the analysis of data characteristics of GPCR, the paper establisheshuman GPCR family prediction model with hidden markov model (HMM). Howeverthe Baum-Welch algorithm is tend to be trapped into local extreme value. Consideringglobal optimization ability of particle swarm optimization algorithm, the paperintroduces adaptive particle swarm optimization algorithm (MAPSO) with multipleweighting coefficient to the training of HMM to obtain optimized parameters of it.Experimental results show that the model is significantly improved in sensitivity,specificity and accuracy.The paper conducts experiments with five super family of human GPCR fromGPCRDB. Experiments confirm that the model functions well in predicting GPCRfamily classification as expected. The sensitivity, specificity and accuracy of themodel are evaluated with methods of data classification and k-fold cross-validation.K-fold cross-validation indicates the average accuracy of family classification ishigher than previous models. A comparative evaluation between proposed algorithmand previous methods proves the proposed improvement of the model is successful in predicting family classification of GPCR. At the end the paper proposes furtherresearch like dimension descending of HMM in case of GPCR sub-family predictingand some others.