Similarity among Tandem Mass Spectra from Proteomic Experiments:  Detection, Significance, and Utility

Abstract

Liquid chromatography paired with tandem mass spectrometry is a standard technique for identifying peptides from complex protein mixtures. Most fragment ion spectra acquired by this technique are unique, but some are repeated. Similarities among the spectra from 1D and 2D liquid chromatography experiments were calculated by the dot product algorithm. Similar spectra were grouped, and the degree of duplication was calculated for each sample. In 1D liquid chromatography data from 1D gel bands, 18% of the fragment ion spectra were duplicates. A six-cycle 2D liquid chromatographic separation of more than 200 proteins produced 28% duplicate spectra. A rat hippocampal homogenate analyzed by a 12-cycle 2D liquid chromatographic separation contained 25% duplicate spectra. Removal of these duplicate spectra, however, resulted in fewer peptides being successfully identified by SEQUEST. We propose a modification for peptide identification algorithms that would improve their performance and accuracy by explicitly recognizing and making use of spectral similarity.

MS2Grouper: Group assessment and synthetic replacement of duplicate proteomic tandem mass spectra

“…The software reads in all spectra from a separation, processes them, and then writes a new set of spectra°to°disk.°The°MS2°file°format° [17]°allows°mark-ups to note which spectra have been synthesized from collections of others, enabling users to trace back to the spectra that gave rise to synthetics. Similarity detection in MS2Grouper is quite like the process described for NoDupe° [8].°The°sum°of°fragment°ion°intensities°for each spectrum is computed. Each fragment ion that contributes less than 1% of this sum is removed from consideration.…”

“…Previous work has shown that this spectral redundancy ranges from 18% for LC-MS/MS analysis of bands from 1D PAGE separation, to 25% for MudPIT proteomic analysis, to 28% for MudPIT analysis of a ϳ200 protein mixture [8]. This phenomenon can be exploited for several purposes.…”

“…LIBQUEST, one of the first algorithms published for comparing experimental peptide spectra to libraries of identified spectra, employed a variation of SEQUEST's cross correlation for comparison [14]. Faster approaches based on dot product comparisons have been more broadly used for spectrum to spectrum comparisons [8,9,11,13]. These algorithms treat individual spectra as vectors in multidimensional space, where the coordinate in each dimension is the intensity at a particular m/z value.…”

“…Computing the cosine of the angle between two such vectors gives a measurement of the similarity of two spectra. Both Tabb et al [8] and Beer et al [13] have reported the use of dot productbased spectral comparisons, specifically for the detection of similar peptide MS/MS spectra in shotgun proteomics data. Where systems like LIBQUEST attempt to match each experimental spectrum to an entry in a library of identified spectra, these latter systems attempt to compare each experimental spectrum to other experimental spectra with similar precursor ion m/z values.…”

“…Tabb et al created "NoDupe" to reveal the amount of similarity present within individual LC-MS/MS analyses across a variety of different experimental models, showing that choosing an arbitrary representative from each group resulted in a minor loss of protein identifications [8]. Beer et al extended the comparison across multiple LC-MS/MS experiments, pairing a system to cluster similar spectra with an algorithm to re-centroid fragment ions observed in replicate spectra to construct a single representative spectrum [13].°By identifying spectral duplication before identification, these systems can improve the efficiency of subsequent database searches.…”