Unravelling associations between unassigned mass spectrometry peaks with frequent itemset mining techniques.

BACKGROUND: Mass spectrometry-based proteomics experiments generate spectra that are rich in information. Often only a fraction of this information is used for peptide/protein identification, whereas a significant proportion of the peaks in a spectrum remain unexplained. In this paper we explore how a specific class of data mining techniques termed "frequent itemset mining" can be employed to discover patterns in the unassigned data, and how such patterns can help us interpret the origin of the unexpected/unexplained peaks. RESULTS: First a model is proposed that describes the origin of the observed peaks in a mass spectrum. For this purpose we use the classical correlative database search algorithm. Peaks that support a positive identification of the spectrum are termed explained peaks. Next, frequent itemset mining techniques are introduced to infer which unexplained peaks are associated in a spectrum. The method is validated on two types of experimental proteomic data. First, peptide mass fingerprint data is analyzed to explain the unassigned peaks in a full scan mass spectrum. Interestingly, a large numbers of experimental spectra reveals several highly frequent unexplained masses, and pattern mining on these frequent masses demonstrates that subsets of these peaks frequently co-occur. Further evaluation shows that several of these co-occurring peaks indeed have a known common origin, and other patterns are promising hypothesis generators for further analysis. Second, the proposed methodology is validated on tandem mass spectrometral data using a public spectral library, where associations within the mass differences of unassigned peaks and peptide modifications are explored. The investigation of the found patterns illustrates that meaningful patterns can be discovered that can be explained by features of the employed technology and found modifications. CONCLUSIONS: This simple approach offers opportunities to monitor accumulating unexplained mass spectrometry data for emerging new patterns, with possible applications for the development of mass exclusion lists, for the refinement of quality control strategies and for a further interpretation of unexplained spectral peaks in mass spectrometry and tandem mass spectrometry.