Detection of complex post-translational modifications (PTMs) remains a challenging open problem in proteomics. Simple PTMs (e.g. methylation, deamidation etc.) can be readily identified in tandem mass spectrometry (MS/MS) by considering characteristic mass shifts in peptide fragment ions. However, more complex PTMs such as glycosylation and small ubiquitin-like modification (SUMOylation) present a more difficult problem because the modification changes the fragmentation pattern of the substrate peptide substantially, making current database search methods not suitable to identify MS/MS spectra from these modified peptides.

We propose a novel approach to enable the expedited development of new algorithms for any types of PTM peptide fragmentation. Using SUMOylation as an example, we demonstrate how to generate large and reliable MS/MS training data from SUMOylated peptides and derive algorithms, Specialize, that learn PTM-specific fragmentation from the training data and apply appropriate false discovery rate (FDR) procedures. Benchmarking our new methods on several datasets of varying complexity from SUMOylation studies on we show that our methods are significantly more sensitive than current state-of-the art methods.