MS2Rescore: Data-Driven Rescoring Dramatically Boosts Immunopeptide Identification Rates

Abstract

•MS2Rescore significantly boosts immunopeptide identification rates•Data-driven post-processing allows for a ten-fold increase in specificity•MS2PIP and DeepLC predictors are integrated with Percolator post-processing•MS2Rescore accepts identification results from MaxQuant, PEAKS, MS-GF+ and X!Tandem•MS2Rescore shows great promise to extend current neo- and xeno-epitope landscapes Immunopeptidomics aims to identify major histocompatibility complex (MHC)-presented peptides on almost all cells that can be used in anti-cancer vaccine development. However, existing immunopeptidomics data analysis pipelines suffer from the nontryptic nature of immunopeptides, complicating their identification. Previously, peak intensity predictions by MS2PIP and retention time predictions by DeepLC have been shown to improve tryptic peptide identifications when rescoring peptide-spectrum matches with Percolator. However, as MS2PIP was tailored toward tryptic peptides, we have here retrained MS2PIP to include nontryptic peptides. Interestingly, the new models not only greatly improve predictions for immunopeptides but also yield further improvements for tryptic peptides. We show that the integration of new MS2PIP models, DeepLC, and Percolator in one software package, MS2Rescore, increases spectrum identification rate and unique identified peptides with 46% and 36% compared to standard Percolator rescoring at 1% FDR. Moreover, MS2Rescore also outperforms the current state-of-the-art in immunopeptide-specific identification approaches. Altogether, MS2Rescore thus allows substantially improved identification of novel epitopes from existing immunopeptidomics workflows. Immunopeptidomics aims to identify major histocompatibility complex (MHC)-presented peptides on almost all cells that can be used in anti-cancer vaccine development. However, existing immunopeptidomics data analysis pipelines suffer from the nontryptic nature of immunopeptides, complicating their identification. Previously, peak intensity predictions by MS2PIP and retention time predictions by DeepLC have been shown to improve tryptic peptide identifications when rescoring peptide-spectrum matches with Percolator. However, as MS2PIP was tailored toward tryptic peptides, we have here retrained MS2PIP to include nontryptic peptides. Interestingly, the new models not only greatly improve predictions for immunopeptides but also yield further improvements for tryptic peptides. We show that the integration of new MS2PIP models, DeepLC, and Percolator in one software package, MS2Rescore, increases spectrum identification rate and unique identified peptides with 46% and 36% compared to standard Percolator rescoring at 1% FDR. Moreover, MS2Rescore also outperforms the current state-of-the-art in immunopeptide-specific identification approaches. Altogether, MS2Rescore thus allows substantially improved identification of novel epitopes from existing immunopeptidomics workflows. The immune system is a complex, yet remarkable system that protects us from both invaders from outside the body, that is, pathogens, as well as from inside the body, that is, malignancies (1Sattler S. Advances in Experimental Medicine and Biology. Springer New York LLC), New York2017: 3-14Google Scholar). Increased understanding of the immune system allowed for great medical achievements such as vaccination, which is currently available for over 29 diseases, enabled the eradication of smallpox, and prevents over 3 million deaths each year (https://www.cdc.gov/vaccines/vpd/vaccines-diseases.html?CDC_AA_refVal=https%3A%2F%2Fwww.cdc.gov%2Fvaccines%2Fvpd-vac%2Fdefault.htm). However, many diseases such as Mycobacterium tuberculosis or malignancies lack effective vaccines due to improper T-cell activation. A key issue in developing effective vaccines for these diseases is the lack of accurately identified major histocompatibility complex (MHC)-presented epitopes or immunopeptides. These epitopes are presented on the cell surface and enable T-cells to discern healthy cells from infected or malignant cells. While much effort has recently been invested in accurate prediction of these epitopes in silico (2Raoufi E. Hemmati M. Eftekhari S. Khaksaran K. Mahmodi Z. Farajollahi M.M. et al.Epitope prediction by novel immunoinformatics approach: a state-of-the-art Review.Int. J. Pept. Res. Ther. 2020; 26: 1155-1163Crossref PubMed Scopus (49) Google Scholar), these are mostly limited to viruses as these contain fewer potential protein antigens (3Mayer R.L. Impens F. Immunopeptidomics for next-generation bacterial vaccine development.Trends Microbiol. 2021; 29: 1034-1045Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar). Moreover, these tools are not yet sufficiently precise to confidently identify epitopes (4Larsen M.V. Lundegaard C. Lamberth K. Buus S. Lund O. Nielsen M. Large-scale validation of methods for cytotoxic T-lymphocyte epitope prediction.BMC Bioinformatics. 2007; 8: 1-12Crossref PubMed Scopus (591) Google Scholar, 5Zhang H. Lundegaard C. Nielsen M. Pan-specific MHC class I predictors: a benchmark of HLA class I pan-specific prediction methods.Bioinformatics. 2009; 25: 83-89Crossref PubMed Scopus (91) Google Scholar). Therefore, experimental immunopeptidomics workflows, such as epitope detection through LC-MS, are still the best way to accurately identify these immunopeptides (6Bassani-Sternberg M. Pletscher-Frankild S. Jensen L.J. Mann M. Mass spectrometry of human leukocyte antigen class i peptidomes reveals strong effects of protein abundance and turnover on antigen presentation.Mol. Cell Proteomics. 2015; 14: 658-673Abstract Full Text Full Text PDF PubMed Scopus (308) Google Scholar). While immunopeptidomics workflows have been readily developed and applied (7Solleder M. Guillaume P. Racle J. Michaux J. Pak H.S. Müller M. et al.Mass spectrometry based immunopeptidomics leads to robust predictions of phosphorylated HLA class I ligands.Mol. Cell Proteomics. 2020; 19: 390-404Abstract Full Text Full Text PDF PubMed Scopus (32) Google Scholar), acquisition of immunopeptides through LC-MS suffers from some major problems. First, the acquisition of immunopeptide spectra is hampered due to the low abundance of immunopeptides and even more so of neo-epitopes. Infrequently occurring epitopes are still very challenging to identify through LC-MS, despite enrichment efforts and sample preprocessing (8Faridi P. Purcell A.W. Croft N.P. In Immunopeptidomics we need a sniper instead of a shotgun.Proteomics. 2018; 18e1700464Crossref PubMed Scopus (51) Google Scholar). Second, in contrast to standard proteomics experiments where proteins are usually digested with trypsin before LC-MS, immunopeptides are captured through immune purification with antibodies followed by acidic elution resulting in mostly nontryptic peptides. The nontryptic nature of immunopeptides results in one less positive charge due to the missing arginine or lysine at the peptide’s C-terminus, causing many immunopeptides to be singly charged during MS acquisition. These singly charged peptides are much harder to analyze because, during fragmentation of the peptide, the charge resides on one of the fragments, leaving the other uncharged and thus lost (9Pfammatter S. Bonneil E. Lanoix J. Vincent K. Hardy M.-P.P. Courcelles M. et al.Extending the comprehensiveness of immunopeptidome analyses using isobaric peptide labeling.Anal. Chem. 2020; 92: 9194-9204Crossref PubMed Scopus (34) Google Scholar). Moreover, most contaminants are singly charged as well, making identifications of immunopeptides much harder (10Purcell A.W. Ramarathinam S.H. Ternette N. Mass spectrometry–based identification of MHC-bound peptides for immunopeptidomics.Nat. Protoc. 2019; 14: 1687-1707Crossref PubMed Scopus (159) Google Scholar). The nontryptic nature of immunopeptides hampers not only the acquisition but also the identification of immunopeptide spectra. To match each acquired spectrum with the peptide from which the spectrum originated, proteomics database search engines such as SEQUEST (11Eng J.K. McCormack A.L. Yates J.R. An approach to correlate tandem mass spectral data of peptides with amino acid sequences in a protein database.J. Am. Soc. Mass Spectrom. 1994; 5: 976-989Crossref PubMed Scopus (5582) Google Scholar), X!Tandem (12Craig R. Beavis R.C. TANDEM: matching proteins with tandem mass PubMed Scopus Google Scholar), J. N. Mann M. a peptide search integrated the Res. PubMed Scopus Google Scholar), or J. M. et database search for and accurate peptide Cell Proteomics. Full Text Full Text PDF Scopus Google in silico spectra for all matching peptides. The of peptides that be in the sample is the search is to that spectra from peptides that are not in search be even In standard proteomics in silico tryptic of proteins a yet search In the search to be of due to from the protein of the of MHC class I peptides, to amino and peptides, to amino J. K. Advances in Experimental Medicine and Biology. Springer New York New and the potential of and immunopeptides, which are peptides that from the or P. C. Ramarathinam S.H. et of peptides are not for and peptide 2018; PubMed Scopus Google Scholar). and protein sequences are as well, even further the search a search leads to more peptide-spectrum matches N. S. K. of the of peptide identification Res. PubMed Scopus Google Scholar), rate and in fewer identified immunopeptides K. H. H. M. and of database search of mass spectrometry based Spectrom. 2020; PubMed Scopus Google Scholar). tryptic peptides have been the standard in search engines as well as tools that in LC-MS spectra are tailored toward tryptic peptides, making less accurate or not at all for The need for neo- and xeno-epitope to the of many tools to improve or identifications in the one tools have been developed that of immunopeptides to immunopeptide the other pipelines have been developed to improve immunopeptide identification. M. C. et and data analysis for Res. 2019; PubMed Scopus Google Scholar), which is a for and J. M. et database search for and accurate peptide Cell Proteomics. Full Text Full Text PDF Scopus Google Scholar). is not for immunopeptides, is due to database which to well for search these tools can with immunopeptide not all available such as retention time and intensity Previously, has been that retention time predictions in standard proteomics workflows can improve identification S. S. K. peptide identifications by spectra identification and retention time Res. 2018; PubMed Scopus Google Scholar). peak intensity predictions to tools such as Percolator can also improve identification R. S. peptide fragmentation predictions data to and improve proteomics search 2019; Scopus Google Scholar), which has been to for immunopeptides as well by efforts such as K. to improve peptide identification in 2020; Scholar, M. M. S. K. et boosts of mass immunopeptidomics.Nat. 2021; PubMed Scopus Google Scholar). tools such as DeepLC R. R. N. S. DeepLC can retention for peptides that 2021; Google and MS2PIP R. S. MS2PIP and accurate peak intensity prediction for fragmentation and Res. 2019; PubMed Scopus Google Scholar, S. MS2PIP prediction and peak intensity predictions for and Res. 2015; PubMed Scopus Google Scholar, S. a for peak intensity 29: PubMed Scopus Google can accurate retention time predictions and peak intensity to in when with identification at a have been to substantially increase R. S. peptide fragmentation predictions data to and improve proteomics search 2019; Scopus Google Scholar). However, DeepLC and MS2PIP are on tryptic peptides. of lysine and arginine at the is less of a for DeepLC as the on retention time is and is for through M.V. Hardy J. Bonneil C. Courcelles M. et proteins the or 2020; Full Text Full Text PDF Scopus Google Scholar). However, is not the for as in peptide as well as fragmentation and methods peak intensity R. S. MS2PIP and accurate peak intensity prediction for fragmentation and Res. 2019; PubMed Scopus Google Scholar). Therefore, we here greatly improved MS2PIP models that include immunopeptides and nontryptic peptides in Moreover, we have integrated MS2PIP and DeepLC with Percolator the and MS2Rescore software which improved rescoring of peptide identifications from proteomics search Altogether, we show that fragmentation spectrum and retention time predictions integrated MS2Rescore increase immunopeptide identification and existing To and new MS2PIP models, available immunopeptide data and one available data from H. P. M. C. et the proteomics identifications 5: PubMed Scopus Google Scholar, J. M. S. et database and tools and in for Res. 2019; PubMed Scopus Google Scholar). for on data a data immunopeptides, a data immunopeptides, the data with tryptic peptides that was used to the existing MS2PIP models, and a data peptide The as well as the of unique peptides and HLA for each data are in tandem mass spectrometry fragmentation are on the fragmentation and applied methods R. S. MS2PIP and accurate peak intensity prediction for fragmentation and Res. 2019; PubMed Scopus Google Scholar), all MS2PIP and data from experiments with the experimental data from was as the models and of data is readily available on each the mass spectrometry to using N. J. M. H. et and Res. 2020; 19: PubMed Scopus Google Scholar). The identification to MS2PIP using and further to unique of peptide and charge at 1% FDR. all spectra one The spectrum J. S. et spectrum for mass 2021; PubMed Scopus Google was used as unique for each to and a peptide identifications and spectra. from each was used as data or as data to data for the where the was used to both data and data was unique to in both to the MS2PIP models, new models with the C. of the on and New Scholar). The J. on Scholar), was used in with for and at for each data as well as for and are shown in To each the was and and peak for each The further by peptide and models the immunopeptide on immunopeptides, the on immunopeptides with peptides, and the nontryptic immunopeptide on nontryptic immunopeptides. models integrated the immunopeptide and the The can be used for immunopeptide peak intensity predictions and the for tryptic and more nontryptic peptide further analysis rescoring immunopeptide only the immunopeptide was as the best for both and immunopeptides. To MS2PIP predictions with the data used as was from and the and models from MS2PIP predictions acquired for the proteomics and data with the and for the and data with the immunopeptide that not in the of the MS2PIP The was in both and spectral to a for singly charged as the MS2PIP models only for these To the of the new MS2PIP models to improve immunopeptide identification the new models with DeepLC and Percolator MS2Rescore based on the search the and the retention and the and the peak These are to Percolator for based on the by et S. E. and accurate database with Res. PubMed Scopus Google and for with search results S. J. The for mass Protoc. PubMed Scopus Google Scholar). MS2PIP used as by et R. S. peptide fragmentation predictions data to and improve proteomics search 2019; Scopus Google Scholar). by MS2Rescore are in MS2Rescore was on a HLA class I data S. S. H. et HLA class I epitope prediction most of the human 2020; PubMed Scopus Google Scholar), which was also used to the recently effort for immunopeptides M. M. S. K. et boosts of mass immunopeptidomics.Nat. 2021; PubMed Scopus Google Scholar). allows both of the improved identification due to the new MS2PIP models and a with First, the identification for the and the Percolator and the mass spectrometry from The mass spectrometry further with N. J. M. H. et and Res. 2020; 19: PubMed Scopus Google and the for each of the rescoring methods using only search a Percolator and using the MS2Rescore search and these rescoring methods compared with the results and with the rescoring was at in of identification rate and of unique identified peptides. The of the in MS2Rescore was using and the of retention time and MS2PIP prediction compared and as by et M. M. S. K. et boosts of mass immunopeptidomics.Nat. 2021; PubMed Scopus Google Scholar), for HLA further with M. Nielsen M. and of peptide Res. PubMed Scopus Google Scholar), for the and lost peptides compared to rescoring with only search To further analyze MS2Rescore for experimental on cells at of and The resulting spectra with the search the human database A peptide of amino was was as with a of Mass at and for and was at with the of a for rescoring with only search and the MS2Rescore for all at all from the to in the of MS2Rescore low and peptides. To MS2Rescore for HLA class peptides, of mass spectrometry from the search results at the spectra with J. M. et database search for and accurate peptide Cell Proteomics. Full Text Full Text PDF Scopus Google with the search that used in the that is, of of with and as the database The identification as well as the from and with MS2Rescore with only search and the MS2Rescore as for the on HLA class I peptides. in using with the a 2007; Scopus Google Scholar), M. data 2021; Google Scholar), and a for mass spectrometry data and Chem. 2020; 92: PubMed Scopus Google In to improve the identification rate of immunopeptides by peak intensity new models for MS2PIP for immunopeptides. using all models improve predictions for both and data in with the tryptic even for standard tryptic proteomics the predictions from the new models are due to the of tryptic peptides the immunopeptide when these peptides are of the in with the While both immunopeptide models are well to peak for tryptic and immunopeptides, the on peptides is not as even and peptides are both are still very for MS2PIP peak intensity immunopeptide peak intensity predictions are improved by all the models, with the immunopeptide the of The are in of a prediction with with the immunopeptide and the less prediction are shown in and C. the of these models is to improve immunopeptide identification by more accurate peak intensity Therefore, identification results rescoring with search a Percolator rescoring with MS2Rescore DeepLC and the new MS2PIP and rescoring with the recently models compared in of the of identifications as well as the of unique identifications based on rescoring with both MS2Rescore and substantially improved the spectrum identification rate in with rescoring with search or not rescoring and at both 1% and FDR. MS2Rescore identification rate of of million compared to for rescoring increase of and only for the search all at 1% A and Moreover, of the identified spectra at 1% are when the to FDR. peak and retention time predictions to Percolator substantially increases the of identified immunopeptides. is by the Percolator for each as well as the for search MS2PIP and DeepLC the of unique identified immunopeptides increases by 36% when MS2PIP and DeepLC for the 1% and even more so for where the of unique identified peptides of the of Percolator identification results These are all HLA class I in the data that the MS2PIP and MS2Rescore, is HLA In MS2Rescore allows for a increase in identification rate for HLA with fewer identifications that MS2Rescore the peptide identification for HLA The of these predictions to Percolator is further when the for and the for and are for both the retention time as well as the the low retention time and A and The are from the and using only the and retention time and both correlate with the search a of and can only be from the by also or retention time and Percolator and when with peak intensity and retention time prediction that have been a 1% of a search are due to a low a retention time or most for the of identified peptides that are lost The integration of and search in MS2Rescore has to substantially increase the identification rate of immunopeptides and outperforms the recently M. M. S. K. et boosts of mass immunopeptidomics.Nat. 2021; PubMed Scopus Google Scholar). In with MS2Rescore and more identifications at 1% and for the of unique identified peptides with a increase of and over unique peptides identified the by MS2Rescore, rescoring only identified these peptides at the 1% the in in of unique identified immunopeptides MS2Rescore thus substantially increases the identification for more for which peak that is, and of the is MS2Rescore more in the data at for the sample A and a these of the MS2Rescore as However, to have a on the of identifications at 1% and and and thus the in identification be to these peptides. a of the peptide spectrum prediction of the MS2PIP models and that of MS2Rescore be to improved peak intensity on the that is MS2PIP or the other on the data is to that for the be for To the in rescoring MS2Rescore and is by a in all MS2Rescore that not have a in rescoring for a all DeepLC retention time and search MS2Rescore and both include peak intensity prediction these the of MS2Rescore is for rescoring which that the retention time and search in improved MS2Rescore over to the identified for HLA was to the in the of the data M. M. S. K. et boosts of mass immunopeptidomics.Nat. 2021; PubMed Scopus Google Scholar, S. S. H. et HLA class I epitope prediction most of the human 2020; PubMed Scopus Google Scholar), the peptides that by MS2Rescore compared to search rescoring less MS2PIP not for in MS2PIP and MS2Rescore be toward spectra with Therefore, search results of mass spectrometry with from to with each MS2Rescore shows a increase in identification However, for the identification rate is most due to a in fragmentation which is in the current and in with the and MS2PIP for when using the in unique identified peptides for MS2Rescore increases for less a increase for by the from fragmentation in of DeepLC retention time and MS2Rescore is to peptides that be lost due to fragmentation spectra. A is for low peptides. the by MS2Rescore in of of unique identified peptides is for the A and where rescoring to most MS2Rescore is thus not only to increase the of identifications for immunopeptides in can peptides lost due to low and thus spectra or To the of MS2Rescore on HLA class peptides, available data was However, for the HLA class I data human immunopeptides and search J. N. Mann M. a peptide search integrated the Res. PubMed Scopus Google results for HLA class data data was with J. M. et database search for and accurate peptide Cell Proteomics. Full Text Full Text PDF Scopus Google Scholar). was the for HLA class I peptides, MS2Rescore significantly increases the identification rate for HLA class peptides with and for the 1% and These increases for the HLA class I data Moreover, where rescoring a increase in to search the in with rescoring is for both identification rate as well as of unique identified peptides. is due to the less search that are for the in MS2Rescore and to the is to identify immunopeptides due to database search J. M. et database search for and accurate peptide Cell Proteomics. Full Text Full Text PDF Scopus Google Scholar). the MS2Rescore peak intensity and retention time still results in a significantly identification Altogether, these results show that MS2Rescore well HLA class I and class immunopeptides, and can the from search new peak intensity prediction models, we to greatly immunopeptide identification rate through While all MS2PIP models greatly peak intensity predictions for immunopeptides, the on immunopeptides the the immunopeptide the of the peptides the not peptides in the data in for the peptides. immunopeptides are much and a charge as a these immunopeptide-specific MS2PIP models are not to the of and charged peptides in the mass While both and peptides are their can be very to of peak intensity of MS2PIP when applied on a of nontryptic peptides. and peptides are immunopeptides and tryptic peptides in of as as almost of the immunopeptide data of tryptic peptides. are thus not However, the of tryptic peptides in immunopeptidomics is most much tryptic most from the tryptic in current immunopeptidomics workflows. in tryptic MHC peptide to R. F. S. J. K. et for the identification of MHC class i Chem. 2018; PubMed Scopus Google Scholar). nontryptic models of we a in tryptic to be to analyze immunopeptide Moreover, by the new immunopeptide with retention time predictions and search MS2Rescore, we greatly the of Percolator to immunopeptide resulting in a immunopeptide identification rescoring increases the of unique identified peptides, which is of for the of potential for or for and to a Moreover, almost identifications at a more MS2Rescore allows a of the to of the peptides identified at 1% FDR. the increase in of the identified MS2Rescore the increase in both and identification MS2Rescore has shown to be with to HLA and the identification by MS2Rescore is even for HLA that fewer that MS2Rescore is to increase the on the immunopeptide for HLA Moreover, MS2Rescore is to peptide identifications that have been lost due to spectra by making of DeepLC retention time and can identifications for low peptides. the of neo- or that less in the MS2Rescore is to immunopeptide identifications of the search for both HLA class I and class peptides, and MS2Rescore with DeepLC and the new immunopeptide MS2PIP models shows improved identification rate over the recently for has shown to more accurate predictions compared to MS2PIP models, is that can be to peak intensity the peak intensity prediction of the new MS2PIP models and of are for immunopeptides even when has been for the These in peak intensity prediction are not the for the of MS2Rescore in of is more that the in rescoring is the of the of more DeepLC, and search when the of the search and all DeepLC retention time the more limited the of MS2Rescore as well a more MS2Rescore a unique that allows Percolator to from identifications much when with limited retention time or peak intensity The of all these is to be of MS2Rescore is available the on and can be through the as well as with a a and a are identification from search engines are and both MS2PIP and DeepLC can a of the need to identification before Altogether, these new models show great promise to greatly extend the immunopeptide in existing and immunopeptidomics MS2Rescore is available at used in is available at The data used for and of the models, the models as well as the MS2Rescore is available on at mass spectrometry proteomics data have been to the the J. M. S. et database and tools and in for Res. 2019; PubMed Scopus Google with the data data (6Bassani-Sternberg M. Pletscher-Frankild S. Jensen L.J. Mann M. Mass spectrometry of human leukocyte antigen class i peptidomes reveals strong effects of protein abundance and turnover on antigen presentation.Mol. Cell Proteomics. 2015; 14: 658-673Abstract Full Text Full Text PDF PubMed Scopus (308) Google Scholar, S. S. H. et HLA class I epitope prediction most of the human 2020; PubMed Scopus Google Scholar, J. Michaux J. M. S. C. et prediction of HLA class epitopes by of 2019; PubMed Scopus Google Scholar, C. F. Pak H. Racle J. M. et and immunopeptidomics reveals of the human leukocyte antigen Proteomics. 2018; Full Text Full Text PDF PubMed Scopus Google Scholar, Guillaume P. Michaux J. Pak Racle J. et and of presented 2018; PubMed Scopus Google Scholar, M. E. R. P. S. et identification of presented on human by mass Scopus Google Scholar, et and abundance of 29 healthy human 2019; Scopus Google Scholar, M. C. Guillaume P. M. Pak H.S. et HLA peptidomes predictions and HLA Scopus Google Scholar, F. Michaux J. Pak H.S. Müller M. et of HLA in immune cells reveals in peptide 2020; Scopus Google Scholar, F. H.S. et analysis of Proteomics. 2018; PubMed Scopus Google Scholar). The that have of with the of the by the R. from the R. from the S. and M. from the M. from the and from from the

Topics & Concepts