Feature-matching Pattern-based Support Vector Machines for Robust Peptide Mass Fingerprinting

Peptide mass fingerprinting, regardless of becoming complementary to tandem mass spectrometry for protein identification, is still the subject of in-depth study because of its higher sample throughput, higher level of specificity for single peptides and lower level of sensitivity to unexpected post-translational modifications compared with tandem mass spectrometry. In this study, we propose, implement and evaluate a uniform approach using support vector machines to incorporate individual concepts and conclusions for accurate PMF.We focus on the inherent attributes and critical issues of the theoretical spectrum (peptides), the experimental spectrum (peaks) and spectrum (masses) alignment. Eighty-one feature-matching patterns derived from cleavage type, uniqueness and variable masses of theoretical peptides together with the intensity rank of experimental peaks were proposed to characterize the matching profile of the peptide mass fingerprinting procedure. We developed a new strategy including the participation of matched peak intensity redistribution to handle shared peak intensities and 440 parameters were generated to digitalize each feature-matching pattern. A high performance for an evaluation data set of 137 items was finally achieved by the optimal multi-criteria support vector machines approach, with 491 final features out of a feature vector of 35,640 normalized features through cross training and validating a publicly available “gold standard” peptide mass fingerprinting data set of 1733 items. Compared with the Mascot, MS-Fit, ProFound and Aldente algorithms commonly used for MS-based protein identification, the feature-matching patterns algorithm has a greater ability to clearly separate correct identifications and random matches with the highest values for sensitivity (82%), precision (97%) and F1-measure (89%) of protein identification.Several conclusions reached via this research make general contributions to MS-based protein identification. Firstly, inherent attributes showed comparable or even greater robustness than other explicit. As an inherent attribute of an experimental spectrum, peak intensity should receive considerable attention during protein identification. Secondly, alignment between intense experimental peaks and properly digested, unique or non-modified theoretical peptides is very likely to occur in positive peptide mass fingerprinting. Finally, normalization by several types of harmonic factors, including missed cleavages and mass modification, can make important contributions to the performance of the procedure. Peptide mass fingerprinting, regardless of becoming complementary to tandem mass spectrometry for protein identification, is still the subject of in-depth study because of its higher sample throughput, higher level of specificity for single peptides and lower level of sensitivity to unexpected post-translational modifications compared with tandem mass spectrometry. In this study, we propose, implement and evaluate a uniform approach using support vector machines to incorporate individual concepts and conclusions for accurate PMF. We focus on the inherent attributes and critical issues of the theoretical spectrum (peptides), the experimental spectrum (peaks) and spectrum (masses) alignment. Eighty-one feature-matching patterns derived from cleavage type, uniqueness and variable masses of theoretical peptides together with the intensity rank of experimental peaks were proposed to characterize the matching profile of the peptide mass fingerprinting procedure. We developed a new strategy including the participation of matched peak intensity redistribution to handle shared peak intensities and 440 parameters were generated to digitalize each feature-matching pattern. A high performance for an evaluation data set of 137 items was finally achieved by the optimal multi-criteria support vector machines approach, with 491 final features out of a feature vector of 35,640 normalized features through cross training and validating a publicly available “gold standard” peptide mass fingerprinting data set of 1733 items. Compared with the Mascot, MS-Fit, ProFound and Aldente algorithms commonly used for MS-based protein identification, the feature-matching patterns algorithm has a greater ability to clearly separate correct identifications and random matches with the highest values for sensitivity (82%), precision (97%) and F1-measure (89%) of protein identification. Several conclusions reached via this research make general contributions to MS-based protein identification. Firstly, inherent attributes showed comparable or even greater robustness than other explicit. As an inherent attribute of an experimental spectrum, peak intensity should receive considerable attention during protein identification. Secondly, alignment between intense experimental peaks and properly digested, unique or non-modified theoretical peptides is very likely to occur in positive peptide mass fingerprinting. Finally, normalization by several types of harmonic factors, including missed cleavages and mass modification, can make important contributions to the performance of the procedure. In MS-based proteomics, MS1 or MS2, or even MSn, data for peptides produced by proteolysis are obtained and used for peptide mass fingerprinting (PMF), 1The abbreviations used are:MS/MSTandem mass spectrometryPMFPeptide Mass FingerprintingPFFPeptide Fragment FingerprintingSVMsSupport Vector MachinesTSTheoretical SpectrumESExperimental SpectrumTPTheoretical PeptideEPExperimental PeakPUDPeptide Uniqueness DatabaseMPIRMatched Peak Intensity RedistributionPCTPeptide Cleavage TypePUNPeptide UniquenessPMAPeptide Mass AlteringPIRPeak Intensity RankFMPFeature-matching Pattern. peptide fragment fingerprinting (PFF), and de novo sequencing for qualitative analysis or quantification of proteins. State-of-the-art proteomics has adopted the use of tandem MS (MS/MS) because of its growing usefulness in protein identification (1Krogan N.J. Cagney G. Yu H. Zhong G. Guo X. Ignatchenko A. Li J. Pu S. Datta N. Tikuisis A.P. Punna T. Peregrín-Alvarez J.M. Shales M. Zhang X. Davey M. Robinson M.D. Paccanaro A. Bray J.E. Sheung A. Beattie B. Richards D.P. Canadien V. Lalev A. Mena F. Wong P. Starostine A. Canete M.M. Vlasblom J. Wu S. Orsi C. Collins S.R. Chandran S. Haw R. Rilstone J.J. Gandi K. Thompson N.J. Musso G. St Onge P. Ghanny S. Lam M.H.Y. Butland G. Altaf-Ul A.M. Kanaya S. Shilatifard A. O'Shea E. Weissman J.S. Ingles C.J. Hughes T.R. Parkinson J. Gerstein M. Wodak S.J. Emili A. Greenblatt J.F. Global landscape of protein complexes in the yeast Saccharomyces cerevisiae.Nature. 2006; 440: 637-643Crossref PubMed Scopus (2350) Google Scholar). PMF is used in with the because of the of the between PMF and is the subject of PMF was the commonly used for protein identification and is still in a is and (1Krogan N.J. Cagney G. Yu H. Zhong G. Guo X. Ignatchenko A. Li J. Pu S. Datta N. Tikuisis A.P. Punna T. Peregrín-Alvarez J.M. Shales M. Zhang X. Davey M. Robinson M.D. Paccanaro A. Bray J.E. Sheung A. Beattie B. Richards D.P. Canadien V. Lalev A. Mena F. Wong P. Starostine A. Canete M.M. Vlasblom J. Wu S. Orsi C. Collins S.R. Chandran S. Haw R. Rilstone J.J. Gandi K. Thompson N.J. Musso G. St Onge P. Ghanny S. Lam M.H.Y. Butland G. Altaf-Ul A.M. Kanaya S. Shilatifard A. O'Shea E. Weissman J.S. Ingles C.J. Hughes T.R. Parkinson J. Gerstein M. Wodak S.J. Emili A. Greenblatt J.F. Global landscape of protein complexes in the yeast Saccharomyces cerevisiae.Nature. 2006; 440: 637-643Crossref PubMed Scopus (2350) Google P. P. R. M. M. C. S. B. A. V. C. K. M. A.M. M. M. M. T. S. A. T. G. G. G. J.M. B. P. G. of the yeast 2006; 440: PubMed Scopus Google Scholar). mass spectrometry Peptide Mass Peptide Fragment Vector Peptide Peak Peptide Uniqueness Peak Intensity Peptide Cleavage Peptide Uniqueness Peptide Mass Peak Intensity Pattern. As by S. P. peptide mass protein PubMed Scopus Google are of PMF are in to analysis of a single the mass matching and of PMF is than of to unexpected post-translational using peptides in PMF of the can by peptides shared by a of proteins. in can than PMF for the analysis of a single PMF a higher sample than PMF in proteomics research we the of protein identification. this several and developed and to using PMF C. the of peptide mass Mass PubMed Scopus Google and P. for experimental 2006; PubMed Scopus Google of the of PMF a for protein identification. Several parameters and used in PMF in to an C. and peptide identification algorithms using MS for use in PubMed Scopus Google and commonly used PMF are important parameters used in PMF (1Krogan N.J. Cagney G. Yu H. Zhong G. Guo X. Ignatchenko A. Li J. Pu S. Datta N. Tikuisis A.P. Punna T. Peregrín-Alvarez J.M. Shales M. Zhang X. Davey M. Robinson M.D. Paccanaro A. Bray J.E. Sheung A. Beattie B. Richards D.P. Canadien V. Lalev A. Mena F. Wong P. Starostine A. Canete M.M. Vlasblom J. Wu S. Orsi C. Collins S.R. Chandran S. Haw R. Rilstone J.J. Gandi K. Thompson N.J. Musso G. St Onge P. Ghanny S. Lam M.H.Y. Butland G. Altaf-Ul A.M. Kanaya S. Shilatifard A. O'Shea E. Weissman J.S. Ingles C.J. Hughes T.R. Parkinson J. Gerstein M. Wodak S.J. Emili A. Greenblatt J.F. Global landscape of protein complexes in the yeast Saccharomyces cerevisiae.Nature. 2006; 440: 637-643Crossref PubMed Scopus (2350) Google the of peptides P. P. R. M. M. C. S. B. A. V. C. K. M. A.M. M. M. M. T. S. A. T. G. G. G. J.M. B. P. G. of the yeast 2006; 440: PubMed Scopus Google the for mass S. P. peptide mass protein PubMed Scopus Google the of and C. the of peptide mass Mass PubMed Scopus Google the of missed cleavage parameters and in peptide mass and the Mass PubMed Scopus Google a peptide R. use of peptide for protein Mass PubMed Scopus Google and R. M. E. S. C. A. protein identification from peptide mass fingerprinting through a algorithm and an peak PubMed Scopus Google of new spectrum F. P. for accurate of peptide mass and its Mass PubMed Scopus Google mass alignment A. S. protein identification with mass PubMed Scopus Google peak C. Yu Peak for peptide mass PubMed Scopus Google T. M. T. T. A and approach to protein identification by the peptide mass fingerprinting use of Scopus Google A. N. G. and of for protein identification using PMF PubMed Scopus Google peak intensity A. S. Peak intensity in mass A study to support PubMed Scopus Google K. E. Peptide Mass Peak Intensity with PubMed Scopus Google mass J.E. Li X. J. and use of peptide mass in 2006; PubMed Scopus Google of high mass on and in peptide mass PubMed Scopus Google mass R. M. and Mass for Peptide Mass PubMed Scopus Google and a for protein identification by using fingerprinting PubMed Scopus Google has the of PMF identification. algorithms in and and each has and to each commonly used In to evaluate the of a study of performance in of specificity and sensitivity of the G. K. J. M. of algorithms for protein identification from using mass spectrometry PubMed Scopus Google J.S. protein identification by using mass spectrometry PubMed Scopus Google P. of accurate mass in protein identification MS or and PubMed Scopus Google and ProFound for Mass Peptide PubMed Scopus Google Scholar). the performance of and ProFound to of from the set a level of using the In to evaluation of new used to the by each of can through a and in are identification can or by complementary S. A.M. on data on and J. Google for from Mass PubMed Scopus Google Scholar). analysis and developed for a tandem mass approach R. N. by analysis of protein mass PubMed Scopus Google and are adopted for G. J. H. J. M. of mass spectrometry data for PubMed Scopus Google are still in of is the and in algorithms and of this study was to incorporate conclusions and new We and a approach to a of parameters for accurate and PMF. We on the inherent attributes and critical issues of theoretical spectrum (peptides), experimental spectrum and spectrum (masses) alignment. of a or a for the identification was the approach, support vector machines was to a of inherent features for protein identification the of a between and study was a derived from theoretical experimental and mass alignment was to conclusions and new A of feature-matching patterns for the PMF were to features for the a set of 35,640 normalized features vector was to a to cross and publicly available “gold standard” PMF of 1733 items. Finally, the optimal with 491 features achieved a evaluation for PMF of 137 items. A data a training is to the PMF data should of a mass spectrum of the of a single protein has by an an identification, R. M. and Mass for Peptide Mass PubMed Scopus Google Scholar). We using data from data set by M. A. mass for protein Mass PubMed Scopus Google is a data set of in by the use of and by and and for of were to the intense MS and by the data set by P. of and its analysis by of and from a proteomics study of protein the protein P. Zhang J. J. of from an in for PubMed Scopus Google the protein patterns of S. during the of in were in by data set was from was by the identification available by and Aldente M. C. C. Aldente and peptide mass fingerprinting protein identification of the through can from the using from the proteomics were to the of to and peak with peak intensity positive PMF the of each data the proteomics data were to to with A PMF was positive the was and the protein was with the protein by the data set using other a PMF a set of experimental peaks is matched to a of theoretical peptides produced by protein in protein is to out of and of the PMF should than the positive in to and to the each of the positive PMF from the in the were and a positive data set of PMF and a data set of PMF from the proteomics A of was to the positive PMF from the and data to the training set and a validating set the PMF from the data set were to the set of each PMF set are in of each PMF and validating are from the and the data of positive to data is are from the data in a new A PMF with the experimental peak to the theoretical peptide individual of the experimental spectrum the theoretical spectrum and the spectrum (masses) alignment are during the PMF was by and were spectrum alignment and to in experimental spectrum, the theoretical spectrum and the spectrum alignment parameters of and parameters to the In this we the parameters of each experimental spectrum or peak of is a of experimental peak experimental peak has a and an intensity to peak the set of of from is a from this are very and by in the in the of peak intensity can because of the of was proposed to the peak the individual peak intensity is by the of the of the peak intensity in each peak In the inherent attribute was to the rank of a peak intensity or intensity We obtained a new the experimental peak intensity rank a spectrum, by was by of peak of highest intense peak is of matching the matching profile of the intense peaks in the experimental of intensity rank is with and in a matched peak intensity set the matched intensity set and the matched intensity set of matched experimental peaks are matching matching peak intensities and matching peak intensities are harmonic matched peak intensities and matched peak intensities are parameters and are A theoretical spectrum or peptide of is a of theoretical peptides M. peptide has a In the from to the of the theoretical cleavage from to in the protein In other theoretical peptides are in the protein to its from to the of cleavage generated the can cleavage or missed We the missed cleavage theoretical missed cleavage and random missed S.J. of missed cleavage in peptides protein identification in PubMed Scopus Google a on was to missed cleavages with to from We used the available to theoretical missed cleavage from random missed of of of the protein and the in should of mass for theoretical is commonly of and for each theoretical peptide is the of its in the protein A peptide is unique in A uniqueness derived from has to for each matched theoretical peptides are than a protein the protein is likely to a protein than a R. Li R. the to fragment for peptide identification via tandem mass PubMed Scopus Google and A. N. G. and of for protein identification using PMF PubMed Scopus Google used in and In this study, we the of matching to the matching profile of a theoretical matched peptides set matched intensities set and matched intensities set of of matched theoretical peptides are of cleavage in each of cleavage of matched theoretical of of experimental peak intensity matched to cleavage in each of cleavage of matched theoretical of of the experimental peak intensity matched to the cleavage in each of cleavage of matched theoretical of matched peptides matched intensities and matched intensities of matched theoretical peptides are of the of single cleavage of matched theoretical peptides of the experimental peak intensity matched to the single cleavage of matched theoretical peptides of the experimental peak intensity matched to the single cleavage of matched theoretical peptides matching matching peak intensities and matching peak intensities are peak intensity the matching peak intensities and matching peak intensities are the matched peptides parameters can A spectrum alignment is commonly by matching masses with a mass between of A single and matching is a can is matched mass in spectrum the in a than a matching because of the mass a new in matched peak intensity for each is peak intensity is has used to with A. S. Peak intensity in mass A study to support PubMed Scopus Google Scholar). the or theoretical peptides are matched to the experimental peak should the intensity of the In study, a new strategy we matched peak intensity redistribution was used to peak intensity or peak intensity for each matched theoretical of types of cleavage and peptide modification, a peptide several masses during As in are to theoretical peptide masses for a theoretical peptide masses of of the of each we a to each of the theoretical peptide masses for a peptide has a is the of the of its cleavage and peptide we to each approach are in a In a matched peptides several theoretical peptides with very A matching can matched peptide intensity for each matched theoretical peptide is a matching can of matched peptide for or of the values together with other peak intensities are to a new peak intensity rank for each matched R. M. and Mass for Peptide Mass PubMed Scopus Google the of mass obtained from the data set using the for missed cleavages of and was a in the and mass for missed cleavages of and to a of matched mass to the spectrum alignment. is the between matched theoretical peptide mass and experimental peak is the of matched mass is the of matched mass is the of matched mass matched mass set the matched peak intensity set and the matched peak intensity set of mass are the parameters matched peak intensity and intensity with mass are In this study, we the attributes of and to a theoretical peptide and to the experimental peak in attribute has are and the values of an is to and each attribute has and A PMF of a set of mass between theoretical peptides and experimental a each attribute of the theoretical peptide and the experimental peak has and items of for a mass alignment. of of is mass and In an to the matching profile of a PMF in we 440 features for each parameters from are for each the of the of experimental theoretical peptides and mass parameters and are normalized by the of matched experimental the of matched theoretical and the of mass the of parameters from to parameters are normalized by harmonic to the of protein for a spectrum are 35,640 features from of feature-matching by of parameters for feature-matching and by is a is used to we used the C. C. a for support vector available in this A was the the and the were using a In this study, positive were from PMF data and were than positive for a for a to the for a an of in a data set by We of with H. C. to to and with to used in approach are in evaluation the in a new In to the feature can the of is a the of of training the of positive and are and the of the feature is C. with feature in and available and are the of the feature of the positive and data is the feature of the positive and is the feature of the the between the positive and and the the each the the likely this feature is to a a by the the feature with of the feature is to the of the features is than of the the new features are to the the lower is new features are the feature set is feature of the for each we use C. with feature in and available to by use to the training set with features and a the to the validating set the evaluation with features and the evaluation of feature with the highest the of the features are higher than the to the new to the training set with the new features and a new the new to validating set the evaluation the new is than the the new the of the new features and on the new the feature set is feature-matching algorithm was compared with the commonly used ProFound and Aldente A of of the algorithms and were used for and to and the for analysis Zhang for and Scholar). In to an a set of and a set were developed for set was from the data with by analysis of peptides via tandem mass spectrometry the PMF from were parameters are in variable of and the of were because of the sample R. M. and Mass for Peptide Mass PubMed Scopus Google parameters set for the PMF missed cleavage in a new PMF were using evaluation was on the identifications by the PMF other were protein were on the of and were was positive the for a correct protein was the or positive the was the algorithm has its to evaluate the of its for Mass Peptide PubMed Scopus Google Scholar). for each algorithm are in the We use performance in including and to evaluate the identification performance and the of with the the of with the the of and of a approach can by the of for feature is to the of feature and the features for a a feature of items from and 35,640 features was of each feature were and by we obtained a features of 491 in to the of feature out of parameters to 491 features are in of and the important in other protein identification and are used to and mass is an important in the of the new used to spectrum and the matched peaks or peptides to other feature matching patterns of parameters to the in a new out of to the 491 is has feature matching and the experimental peaks with rank intensity to the performance than other of feature matching patterns to the in a new the of attributes and and of attributes were the to the are important of theoretical peptides than missed unique peptides other than peptides a on of mass including and variable modifications a for the intensity experimental peaks a on performance and experimental peaks with and intensity a comparable the of each harmonic are in by harmonic has higher than the is to the derived from in the experimental of harmonic to the in a new in the were on the 491 We used to peptide mass for missed cleavage and mass in an was used to the for each of of cleavage and mass are were to values of for each of were from on the of a of and was used to the and high values of each were set to the from A of of were to and to of cleavage and of mass are We the from of performance on and derived a to the experimental of the was the the of the parameters theoretical missed random missed and variable on the of values for A missed missed and are and for each are in values of each were used in and to evaluate the of is is the highest of performance and parameters were used in the of the of PMF from experimental is a is by mass of and intensity peak mass and the of peaks from a experimental spectrum can between and even between in the peaks a positive on protein identification. and the the algorithm by is to the peak used for protein identification. We used a approach strategy to the peak the peaks of each PMF set in and were and used for the from were to evaluate A of high performance was by of performance evaluation of peak were by between and the highest precision and of the experimental peak was was for the of the the of the intense peaks should to protein identification, a of matching was the of intense peaks on is to the rank of experimental peak we used a approach to the rank of with an were used to and to were used to evaluate A of high performance was by of the evaluation rank were by between and an of precision of and the and of the rank of experimental peak intensity was set the data were with Mascot, MS-Fit, and using a set of parameters and the performance performance of algorithms on the PMF data set of in a new and achieved to by for Mass Peptide PubMed Scopus Google Scholar). the algorithms use the and with of and and of the Aldente of the proteins. algorithm of the by the highest sensitivity (82%), precision and F1-measure (89%) values of protein identification. of are in and ProFound greater ability to clearly separate correct identifications and random matches this was in the of of peptide by the PMF algorithms are in a in of correct from the algorithms were by algorithms were by a single algorithm and of protein were by regardless of becoming complementary to for protein identification, the commonly used in proteomics by in-depth in is to incorporate conclusions and a uniform approach to the performance of protein identification. In this study, types of were adopted parameters parameters derived from matched mass to for the final still a new the PMF procedure. of missed cleavage has in to is to with theoretical and random missed cleavages of matched peptides on a protein for protein identification. to of Peak has by because is by and the specificity for is becoming a of with the between peak intensity and protein intense experimental peaks are likely to generated from peptides than we to peak intensity in this approach by using than peak intensity was used to make comparable between PMF peak intensity and peak intensity in the of of of matching are proposed to the of intense experimental peaks on Finally, parameters derived from peak intensity and peak intensity out of to the Peak intensity is an inherent attribute of the experimental spectrum and should receive attention during protein identification. of is to handle shared peak intensities during spectrum alignment. make to shared peak intensities to peptides of inherent of peptide missed cleavage and mass to of showed missed cleavages should an to protein identification. normalized by several types of harmonic factors, missed cleavages and mass can on is to or and the algorithm to the of the training data is on the of the are by In an to the PMF in we and attributes of the experimental and theoretical spectrum to A of parameters for each the and to positive from of to with the in performance with In other between intense experimental peaks and theoretical peptides from cleavage or unique in the or with the mass are very likely to in positive parameters derived from attributes of the experimental spectrum and the theoretical spectrum to the out of parameters were derived from peak intensity rank and peak intensity of the experimental spectrum, and parameters were derived from the peptide cleavage of the experimental performance of the attributes comparable or even greater robustness than other of and of make a in In this study, was used to and parameters from 35,640 features generated from a of and new 491 features of the PMF we obtained high performance with the validating data set by feature-matching approach algorithm and has a greater ability to from matched the highest precision and F1-measure high level of performance is the of the parameters and features in procedure. As by the of the of experimental is to peaks for single PMF data to this feature-matching approach to by PMF data to its performance by A for protein identification on the approach is available We to of of and for the mass spectrometry data of set and of and for

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.