A New Method For Predicting Protein Structural Classes

In recent years, with the rapid development of the genome sequencing technology and thematerial structural analysis technology, the total amount of data of biological sequences andstructures increases fast. The traditional experimental methods have not meet the needs ofdealing with the huge amount of biological data. In this case, Bioinformatics, whose aim is toreveal the biological significance through gaining, processing, storing, searching and analysingthe biological data, emerges as a new and developing interdiscipline. Recently, Bioinformaticshas been a necessary tool of Life Sciences and Biotechnology.In this thesis, we propose a new method for predicting protein structural classes. The maincontent is as follows：Based on16kinds of classifications of the amino acids, we obtain thederived sequences of a protein sequence. Combining the weighted pseudo-entropy withLempel-Ziv complexity, we construct a34-D feature vector to represent a protein sequence. Thenit is applied to predict the protein structural classes by means of the Bayes classifier. The datasetincludes640sequences that share sequence identity below25%. The accuracy is71.28%. It isshown that our approach represents an improvement in the prediction of accuracy over existingmethods.