A Study On Feature Extraction And Classification Algorithms For Antimicrobial Peptides And Anticancer Peptides

Antimicrobial Peptides(AMPs) represent a class of micro-molecule polypeptides that from biology itself, these peptides are effective for wide range of pathogens and not easy to induce resistance, so they play an important role in innate immune system. Anticancer Peptides(ACPs) are a kind of antimicrobial peptides which has antitumor activity, they kill tumor cells by disrupting membrane or other methods. The antimicrobial peptides research has important significance for the development of new drugs, it has become a hot issue for the scientists to adopt effective methods to predict the antimicrobial peptides and their biological function.In this paper, we chose two datasets about AMPs and ACPs from the published literature, by selecting a variety of feature vector informations, three algorithms(support vector machine,SVM; randomforest,RF,and weighted K-nearest neighbors, WKNN) have been applied to predict these peptides, the prediction efficiency was evaluated by the5-fold cross-validation. In stuty,when using amino acid sequence(AAC) information and pseudo amino acid composition(PseAAC), the forecast effect is better. In the prediction of AMPs,when we chooe amino acid composition as feature vector imformation, the best accuracy is90.83%and4.51%higher than the published literature. In the prediction of ACPs,when we chose the combined vector of amino acid composition and pseudo amino aid composition as feature vector,the best accuracy is93.31%and3.61%higher than the published literature. The result shows that the methods in this paper can predict effectively against AMPs and ACPs.