Searle et al. (2015) show how PREGO, a software tool for predicting high-responding peptides, is useful in creating selected reaction monitoring (SRM) assays for proteomics analysis.1 Following artificial neural network training with data extracted from data-independent acquisition (DIA) tandem mass spectrometry (MS/MS) proteomics analysis, PREGO enabled assay improvements of between 40% and 85% over previous approaches.
SRM is a valuable tool for quantitative proteomics analysis. However, developing the conditions for the focused monitoring environment required for the assay is time-consuming. In order for the assay to run accurately, researchers must first identify unique peptide sequences for the proteins under investigation. Furthermore, these peptide sequences must qualify as high responding in that they will produce a high SRM signal under MS analysis. Once the sequences are identified, researchers must optimize and validate the quantitative workflow with exhaustive testing.
To assist the development process, scientists use various software tools that can help speed up peptide prediction for SRM assay development. These often base selection choices on results acquired using collections of digested protein samples as training sets for machine learning. PREGO, written in Java, also uses training sets, but in this case, these comprise equimolar synthetic peptide mixes that avoid the problems commonly encountered in the digestion process.
Searle and coauthors set up a training set comprising stable isotope labeled (SIL) peptides (n = 1,679) based on tryptic digests of human proteins as observed in data-dependent acquisition (DDA) shotgun proteins runs. They performed liquid chromatography (LC)-MS/MS analysis on a Q Exactive HF mass spectrometer with electrospray ionization (Thermo Scientific), collecting both DDA and DIA spectral data (20 MS/MS scans each). The team processed the DDA data with Comet 2014.02 rev. 2 and peptide spectral matches (PSMs) with Percolator v.2.07. They used Bibliospec v.2.0 to add the PSMs to a spectral library. They analyzed DIA data with Skyline.
Next, the researchers examined the SRM testing set in addition to a training cross-validation set derived from previous studies. They created a synthetic set of more than 700 proteins, then digested them with trypsin. Searle et al. then examined these peptides using a TSQ Vantage triple quadrupole mass spectrometer (Thermo Scientific).
From the DIA training set evaluation, the researchers obtained 1,186 “well-behaved” peptides suitable for assay development. Applying the results from PREGO to create a quantitative SRM assay, Searle et al. found they could achieve an improvement (40–85%) in quantitative proteomics results compared with standard methods for developing SRM assays.
In evaluating PREGO, Searle et al. note that the training approach to fine-tuning the algorithm used for peptide prediction and selection improves on a rules-based approach or random selection. They suggest that although there is an opportunity to improve on initial results, researchers will have better success in building SRM assays by capitalizing on PREGO’s training using the DIA data sets.
1. Searle, B.C., et al. (2015) “Using data independent acquisition to model high-responding peptides for targeted proteomics experiments,” Molecular and Cellular Proteomics, 14 (pp. 2331–2340)
Post Author: Amanda Maxwell. Mixed media artist; blogger and social media communicator; clinical scientist and writer. A digital space explorer, engaging readers by translating complex theories and subjects creatively into everyday language.