Abstract:

In order to effectively deal with the choice blindness of parameters and unbalanced class sizes, TIGA-S4VM, a protein sequence identification model was developed and trained using safe semi-supervised support vector machine (S4VM) based on improved TF-IDF algorithm and Genetic Algorithm (GA). LBTF-IDF, the improved TF-IDF algorithm, was put forward in this model for extracting the protein sequences′  features. After the normalization of features′  frequencies, the results were taken as the characteristic values for classifier. Combining LBTF-IDF, GA and S4VM, the mixed strategy was used to identify the protein sequences. Experiment results showed that the method was superior to other five classification methods and could get good classification performance with reduced training set.

Key words: GA, protein sequence identification, semi-supervised algorithm, S4VM, TF-IDF, support vector machine