Ions classification in peptide tandem mass spectra

In computational proteomics, inferring the peptide sequence from its MS/MS data is an important issue and many algorithms have been proposed recently. Ions classification aiming at determining the type of ions provides a basis for most of the existing algorithms. However, no report on ions classification methods have been found to our knowledge. In this paper, a method extracting ion feature is first presented according to the analysis of the relationship among ions. To deal with ions with high overlap peaks and high density peaks in some mass interval, a method of filtering 'noise' peaks is then proposed according to the information of the related ions. Moreover, a binary ions classification method, which takes some type of ions as one class and the rest ions as the other class, is proposed based on SVM with a novel kernel trick. In the experiments, classification for bions and y-ions are implemented. The results demonstrate that an accuracy level of 90% is achieved.