Gene Classification Using Codon Usage and Support Vector Machines

A novel approach for gene classification, adopting codon usage bias as feature inputs to support vector machines (SVMs) is proposed. The DNA sequence is first converted to a 59-dimensional feature vector, where each element corresponds to the relative synonymous usage (RSCU) frequency of a codon. Since the input to the classifier is independent of sequence length, the approach is especially useful when sequences to be classified are of differing lengths and homology-based methods tend to fail. The method is demonstrated with 1,841 Human Leukocyte Antigen (HLA) sequences, which are classified into two major classes, HLA-I and HLA-II. Each major class is further classified into subgroups. Using codon usage frequencies, binary SVM achieved an accuracy rate of 99.3 percent for HLA major class classification and multiclass SVM achieved accuracy rates of 99.73 percent and 98.38 percent for the subclass classification of HLA-I and HLA-II molecules, respectively. Comparisons with K-Means clustering and other classifiers and homology-based features are given. Results indicate that the classification based on codon usage bias is consistent with biological functions of HLA molecules.