The three-dimensional (3D) structure prediction of proteins is an important task inbioinformatics. Finding energy functions that can better represent residue-residueand residue-solvent interactions is a crucial way to improve the prediction accu-racy. The widely used contact energy functions mostly only consider the contactfrequency between different types of residues; however, we find that the contactfrequency also relates to the residue hydrophobic environment. Accordingly, wepresent an improved contact energy function to integrate the two factors, which canreflect the influence of hydrophobic interaction on the stabilization of protein 3Dstructure more effectively. Furthermore, a fold recognition (threading) approachbased on this energy function is developed. The testing results obtained with 20randomly selected proteins demonstrate that, compared with common contact en-ergy functions, the proposed energy function can improve the accuracy of the foldtemplate prediction from 20% to 50%, and can also improve the accuracy of thesequence-template alignment from 35% to 65%.
A computational system for the prediction and classification of human G-proteincoupled receptors (GPCRs) has been developed based on the support vector ma-chine (SVM) method and protein sequence information. The feature vectors usedto develop the SVM prediction models consist of statistically significant featuresselected from single amino acid, dipeptide, and tripeptide compositions of pro-tein sequences. Furthermore, the length distribution difference between GPCRsand non-GPCRs has also been exploited to improve the prediction performance.The testing results with annotated human protein sequences demonstrate that thissystem can get good performance for both prediction and classification of humanGPCRs.
G-protein coupled receptors (GPCRs) represent one of the most important classesof drug targets for pharmaceutical industry and play important roles in cellularsignal transduction. Predicting the coupling specificity of GPCRs to G-proteins isvital for further understanding the mechanism of signal transduction and the func-tion of the receptors within a cell, which can provide new clues for pharmaceuticalresearch and development. In this study, the features of amino acid compositionsand physiochemical properties of the full-length GPCR sequences have been ana-lyzed and extracted. Based on these features, classifiers have been developed topredict the coupling specificity of GPCRs to G-proteins using support vector ma-chines. The testing results show that this method could obtain better predictionaccuracy.
Many researchers have used microarray gene expression data to investigate gene regulatory networks in specific life stages. In these analyses,Bayesian network was widely applied to regulatory network building from expression profiles because of its solid mathematical foundation and its robust analysis ability in noisy data. However,the building of Bayesian network is time consuming and the searching space is really large. Considering the biological feature of transcription factors (TFs) and targets (TGs),the regulatory network is possible to be separated into core TFs networks and the interactions from TFs to TGs. We developed an R package named ModuleNet which used Bayesian network model to the inner TFs network building and genetic algorithm on TF-TG interactions prediction. With determined number of transcription factors,the searching space and time requirements of ModuleNet is linear increasing according to the number of targets. After application to yeast cell-cycle expression profile,the results demonstrated the prediction accuracy of ModuleNet. Furthermore,significantly enriched Gene Ontology (GO) terms with similar expression behaviors were detected automatically by ModuleNet from expression profile,and the relationships from TFs to GO terms were figured out. The source code is available by asking for the author.
