pepReap: A Peptide Identification Algorithm Using Support Vector Machines

Protein identification plays an important role in proteomics. An algorithm for peptide identification using support vector machines (SVM), pepReap, which consists of two-layered scoring scheme, is designed and implemented. First, a list of peptide candidates is obtained by coarse scoring calculated from total intensity and number of matched peaks, and peptide length. Second, the above preliminary peptide candidates are evaluated by an SVM-based scoring scheme using other important factors, such as correlations between ions, average match error, peptide sequence information, to improve the reliability of peptide identifications. Matthews correlation coefficient is used to measure the classification performance in the SVM training process in order to accommodate to unbalanced datasets. Experiments on a public dataset of tandem mass spectra demonstrate that the pepSeap algorithm outperforms the popular software SEQUEST which uses threshold evaluation in terms of identification sensitivity with comparable precision.