Protein classification using family profiles

Protein classification plays an important role in the research in Bioinformatics. Many discriminative methods, including the SVM based algorithms are used to do this job. In order to use these methods, variable length protein sequences must be converted into fixed-length dimensional vectors. The current work presents a new method of converting sequences into vectors. The method first constructs profile sequences for each protein domain family, then the alignment values of every family profile sequence with a single protein sequence, is used as the protein's according vectors. Then classification algorithms are used to train and predict protein sequences involved. Experiments were presented to test the ability of the SVM algorithm and the LS-StaticEField algorithm to recognize previously unknown sequences via this converting method. Experimental results show that the converting method is good enough and that the LS-StaticEField algorithm is comparable with the SVM one.