Jonathan Krieger (Mississauga / CA), Tharan Srikumar (Mississauga / CA), Dennis Trede (Bremen / DE), George Rosenberger (Faellanden / CH), Arthur Declerq (Ghent / BE), Ralf Gabriels (Ghent / BE), Robbin Bouwmeester (Ghent / BE), Sven Degroeve (Ghent / BE), Lennart Martens (Ghent / BE), Michele Genangeli (Leiderdorp / NL)
Introduction:
The immunoproteome, the subset of protein involved in immune response, are key components of various disease mechanisms and recent progress in MS instrumentation and analysis have gained considerable attention for this special application. However, because immunopeptides are non-enzymatically cleaved, the necessary search space dramatically increases, resulting in lower peptide identification rates. It has been demonstrated that the introduction of orthogonal scores comparing predicted vs observed peptide properties can increase both sensitivity and specificity in these scenarios. Here, we introduce TIMSrescore, a timsTOF-optimized algorithm based on MS2Rescore, that provides these benefits for the timsTOF family of instruments.
Methods & Results:
The peptide fragmentation (MS2PIP) and collisional cross section (im2deep) predictors employed in TIMSrescore have been trained using a diverse set of timsTOF dda-PASEF PSMs. The database search engine Sage and MS2Rescore have been extended to support timsTOF data. For benchmarking, three different types of datasets were used: 1) a single dda-PASEF run of an in-house whole cell lysate proteome mixture (human, yeast, E. coli), 2) a single dda-PASEF run of a fractionated, phosphopeptide-enriched dataset (Skowronek et al., MCP, 2022), 3) a single dda-PASEF run of an HLA-I immunopeptidomic dda-PASEF dataset (Phulphagar et al., MCP, 2023). Sage and MS2Rescore were run with default parameters for standard, phosphoproteomic and immunopeptidomic analyses (8-15 AA).
For the whole cell lysate and phosphoproteomic datasets, TIMSrescore increased recovery of peptides by 2.1 and 2.0%, respectively. However, 12.5 and 15.9% of peptides were reprioritized, due to higher confidence after rescoring. For the immunoproteomic dataset, a considerable increase of 21.8% could be observed, indicating that rescoring is particularly impactful for large search spaces. Particularly the timsTOF optimizations of the fragmentation predictors were critical, improving median Pearson correlation to 0.88 from 0.53 (standard MS2PIP HCD model).
Conclusions:
TIMSrescore represents a timsTOF-optimized rescoring approach that can improve recovery of peptide identifications on immunopeptidomic applications substantially.