The Research Of Prediction Method Of Protein Structural Class Based On Linear Predictive Coding Of PSI-BLAST Profiles

Bioinformatics is a new and popular interdiscipline，which usedmathematics, computer science and biology as its research tool. Theresearch of bioinformatics mainly focused on analyzing and explainingmass molecular biology data information by statistics and computerscience. Bioinformatics has a wide range of research ifeld which almostinvolve in all ifeld life science. Based solely on one protein sequenceinformation to predict protein secondary structural class is a basic andfundamental problem in bioinformatics.In the chapter two, we introduce some popular methods to determineprotein structure. There are three expeirments methods to determineprotein3-D structure and a compute method to predict protein structure.In the steps of compute method to predict protein structure, some simplefeature extraction method and popular machine study classifiers containk-Nearest Neighbors (k-NN) algorithm, Bayesian classifier and SupportVector Machine (SVM).In the chapter three, a new method to predict protein secondarystructural class was proposed. We applied a signal processing tool， namely linear predictive coding (LPC), to extract features and predictstructural class of a protein sequence. First, the PSI-BLAST program wasemployed to transform evolutionary information of the oirginal proteinsequences to position-specific score matrices (PSSM). Then, the LPCalgorithm is applied to extract local correlation features of peptides fromPSSMs of protein sequences, and the final LPC coefifcients are served aspredictive features. Cross-validation tests by support vector machineshowed that the proposed method takes a signiifcant leap at the overallaccuracy on four benchmark datasets. This study provides a new featureextraction method and achieves a good performance on protein structuralclassification problem.