Studies On Several Problems Of Protein Structure Prediction

As the performance of genetic information, proteins are theimportant biological macromolecules. The study found that the spatialstructure of the protein function and protein closely linked, similar to thefunction of the protein structure is generally similar. Therefore, theprediction of protein structure to understand the function of the protein,and thus help to reveal the nature of the activities of life, and theunderstanding of disease mechanisms and targeted drug development willplay a positive role in promoting. With the continuous development ofhigh-throughput sequencing technology, the number of protein sequencesappears a exponential growth trend. Therefore, using the experimentalmethod to obtain the protein structure can’t meet the need. Thus, the useof compute method for protein structure prediction has become a hottopic in the bioinformatics research.Prediction of protein3D structure from solely its amino acidsequence is one of the most challenging problems in structuralbioinformatics, where the3D structure reconstruction from observedconstraints is the key step. In this paper, we propose a novel protocolGlocal to recover a protein’s3D coordinates based on a given2D contact map by combining both global and local optimization schemes achievedby the swarm intelligence of Particle Swarm Optimization (PSO) and theSimulated Annealing (SA) techniques respectively. Our resultsdemonstrate that Glocal can recover the3D structures with the averageRMSD less than2from the native contact map. Further analysis alsoshows that Glocal is particularly powerful for handling with noisy contactmap with the proposed combination optimization approaches.Disulfide bonds, formed by the cysteine pairs, play important roleson stabilizing the protein structure. Disulfide connectivity predictiondepends largely on the feature representation as the input of predictingmodel. In this paper, we combine the correlated mutation-based enginePSICOV and the OET-KNN machine learning-based method. Weimprove traditional machine learning-based method mainly on the featurerepresentation by introduce domain information and propose a parallelframework that will represent samples in complex spaces. Ourexperiment demonstrates the improvement on disulfide connectivityprediction own to these new feature representations. At the other hand,disulfide bond is a well defined residue–residue interaction that plays animportant role for the protein structure and function, so it is expected toresult in the co-evolution of these positions. The correlated mutationanalysis engine is implemented by the recently developed PSICOValgorithm, which calculates residue correlations from multiple sequence alignments. The outputs of the OET-KNN classifier and PSICOV aremerged by using a linear combination to predict disulfide connectivity.The accuracy of the method we developed to predict disulfide bondconnectivity patterns reach as high as79.2%...