Hierarchical Clustering of Shotgun Proteomics Data

Abstract

A new result report for Mascot search results is described. A greedy set cover algorithm is used to create a minimal set of proteins, which is then grouped into families on the basis of shared peptide matches. Protein families with multiple members are represented by dendrograms, generated by hierarchical clustering using the score of the nonshared peptide matches as a distance metric. The peptide matches to the proteins in a family can be compared side by side to assess the experimental evidence for each protein. If the evidence for a particular family member is considered inadequate, the dendrogram can be cut to reduce the number of distinct family members. A new result report for Mascot search results is described. A greedy set cover algorithm is used to create a minimal set of proteins, which is then grouped into families on the basis of shared peptide matches. Protein families with multiple members are represented by dendrograms, generated by hierarchical clustering using the score of the nonshared peptide matches as a distance metric. The peptide matches to the proteins in a family can be compared side by side to assess the experimental evidence for each protein. If the evidence for a particular family member is considered inadequate, the dendrogram can be cut to reduce the number of distinct family members. In shotgun proteomics, a mixture of proteins, which may be as complex as a whole cell lysate, is digested to peptides prior to chromatographic separation and analysis by mass spectrometry. Database searching of the tandem MS (MS/MS) 1The abbreviations used are:MS/MStandem MSBLASTBasic Local Alignment Search ToolFDRfalse discovery rateiTRAQIsobaric Tags for Relative or Absolute Quantitation. spectra delivers matches to peptide sequences. Using these matches to deduce which proteins were present in the original sample is surprisingly difficult because many of the peptide sequences in a typical search result can be assigned to more than one protein. tandem MS Basic Local Alignment Search Tool false discovery rate Isobaric Tags for Relative or Absolute Quantitation. A comprehensive description of the “Protein Inference Problem” can be found in the review by Nesvizhskii and Aebersold (1Nesvizhskii A.I. Aebersold R. Interpretation of shotgun proteomic data - The protein inference problem.Mol. Cell. Proteomics. 2005; 4: 1419-1440Abstract Full Text Full Text PDF PubMed Scopus (795) Google Scholar). More recently, computational tools for protein inference and estimation of protein false discovery rate (FDR) have been reviewed by Li et al. (2Li N. Wu S.F. Zhu Y.P. Yang X.M. The progress of protein quality control methods in shotgun proteomics.Prog. Biochem. Biophys. 2009; 36: 668-675Crossref Scopus (6) Google Scholar), who observed that they can be categorized as deterministic approaches (DBParser (3Yang X. Dondeti V. Dezube R. Maynard D.M. Geer L.Y. Epstein J. Chen X. Markey S.P. Kowalak J.A. DBParser: web-based software for shotgun proteomic data analyses.J. Proteome Res. 2004; 3: 1002-1008Crossref PubMed Scopus (88) Google Scholar), Mass Sieve (4Slotta D.J. McFarland M.A. Markey S.P. MassSieve: Panning MS/MS peptide data for proteins.Proteomics. 2010; 10: 3035-3039Crossref PubMed Scopus (49) Google Scholar), EPIR (5Kristensen D.B. Brønd J.C. Nielsen P.A. Andersen J.R. Sørensen O.T. Jørgensen V. Budin K. Matthiesen J. Venø P. Jespersen H.M. Ahrens C.H. Schandorff S. Ruhoff P.T. Wisniewski J.R. Bennett K.L. Podtelejnikov A.V. Experimental Peptide Identification Repository (EPIR): An integrated peptide-centric platform for validation and mining of tandem mass spectrometry data.Mol. Cell. Proteomics. 2004; 3: 1023-1038Abstract Full Text Full Text PDF PubMed Scopus (34) Google Scholar), Isoform Resolver (6Resing K.A. Meyer-Arendt K. Mendoza A.M. Aveline-Wolf L.D. Jonscher K.R. Pierce K.G. Old W.M. Cheung H.T. Russell S. Wattawa J.L. Goehle G.R. Knight R.D. Ahn N.G. Improving reproducibility and sensitivity in identifying human proteins by shotgun proteomics.Anal. Chem. 2004; 76: 3556-3568Crossref PubMed Scopus (204) Google Scholar), DTASelect (7Tabb D.L. McDonald W.H. Yates 3rd, J.R. DTASelect and contrast: Tools for assembling and comparing protein identifications from shotgun proteomics.J. Proteome Res. 2002; 1: 21-26Crossref PubMed Scopus (1141) Google Scholar), ProteinScape (8Stephan C. Reidegeld K.A. Hamacher M. van Hall A. Marcus K. Taylor C. Jones P. Müller M. Apweiler R. Martens L. Körting G. Chamrad D.C. Thiele H. Blüggel M. Parkinson D. Binz P.A. Lyall A. Meyer H.E. Automated reprocessing pipeline for searching heterogeneous mass spectrometric data of the HUPO brain proteome project pilot phase.Proteomics. 2006; 6: 5015-5029Crossref PubMed Scopus (44) Google Scholar), IDPicker (9Zhang B. Chambers M.C. Tabb D.L. Proteomic parsimony through bipartite graph analysis improves accuracy and transparency.J. Proteome Res. 2007; 6: 3549-3557Crossref PubMed Scopus (264) Google Scholar, 10Ma Z.Q. Dasari S. Chambers M.C. Litton M.D. Sobecki S.M. Zimmerman L.J. Halvey P.J. Schilling B. Drake P.M. Gibson B.W. Tabb D.L. IDPicker 2.0: Improved protein assembly with high discrimination peptide identification Filtering.J. Proteome Res. 2009; 8: 3872-3881Crossref PubMed Scopus (269) Google Scholar), PROVALT (11Weatherly D.B. Atwood 3rd, J.A. Minning T.A. Cavola C. Tarleton R.L. Orlando R. A heuristic method for assigning a false-discovery rate for protein identifications from Mascot database search results.Mol. Cell. Proteomics. 2005; 4: 762-772Abstract Full Text Full Text PDF PubMed Scopus (162) Google Scholar)) or probabilistic approaches (Qscore (12Moore R.E. Young M.K. Lee T.D. Qscore: An algorithm for evaluating SEQUEST database search results.J. Am. Soc. Mass Spectrom. 2002; 13: 378-386Crossref PubMed Scopus (337) Google Scholar), PRISM (13Kislinger T. Rahman K. Radulovic D. Cox B. Rossant J. Emili A. PRISM, a generic large scale proteomic investigation strategy for mammals.Mol. Cell. Proteomics. 2003; 2: 96-106Abstract Full Text Full Text PDF PubMed Scopus (149) Google Scholar), ProteinProphet (14Nesvizhskii A.I. Keller A. Kolker E. Aebersold R. A statistical model for identifying proteins by tandem mass spectrometry.Anal. Chem. 2003; 75: 4646-4658Crossref PubMed Scopus (3635) Google Scholar), PRO_PROBE (15Sadygov R.G. Liu H. Yates J.R. Statistical models for protein validation using tandem mass spectral data and protein amino acid sequence databases.Anal. Chem. 2004; 76: 1664-1671Crossref PubMed Scopus (117) Google Scholar), PANORAMICS (16Feng J. Naiman D.Q. Cooper B. Probability model for assessing proteins assembled from peptide sequences inferred from tandem mass spectrometry data.Anal. Chem. 2007; 79: 3901-3911Crossref PubMed Scopus (57) Google Scholar), and EBP (17Price T.S. Lucitt M.B. Wu W. Austin D.J. Pizarro A. Yocum A.K. Blair I.A. FitzGerald G.A. Grosser T. EBP, a program for protein identification using multiple tandem mass spectrometry datasets.Mol. Cell. Proteomics. 2007; 6: 527-536Abstract Full Text Full Text PDF PubMed Scopus (47) Google Scholar)). A review by Shi and Wu (18Shi J.H. Wu F.X. Protein inference by assembling peptides identified from tandem mass spectra.Curr. Bioinf. 2009; 4: 226-233Crossref Scopus (10) Google Scholar) contains additional discussion of how peptide uniqueness and detectability can be used, such as proteotypic peptides (19Kuster B. Schirle M. Mallick P. Aebersold R. Scoring proteomes with proteotypic peptide probes.Nat. Rev. Mol. Cell Biol. 2005; 6: 577-583Crossref PubMed Scopus (304) Google Scholar) and a Bayesian model that penalizes a protein for the absence of a match to an expected peptide (20Li Y.F. Arnold Li P. H. A Bayesian to protein inference in shotgun proteomics.J. Biol. 2009; PubMed Scopus Google Scholar). approaches protein as a basis for protein identifications that be as such as proteins identified by a peptide J. Zimmerman L.J. Tabb D.L. D.C. B. protein identification and data in shotgun Biol. 2009; Scopus Google spectral in which MS/MS spectra are into then to protein sequences N. P.A. discovery of protein a the Proteome Res. 2009; 8: PubMed Scopus Google the of peptides to a model M.A. E. E. H. R. B. K. Ahrens C.H. protein inference for shotgun data new into and Res. 2009; PubMed Scopus Google Scholar, E. Ahrens C.H. for protein inference and 2010; PubMed Scopus Google and analysis to the for an set of protein identifications L. M. S.P. M. A. Aebersold R. Protein identification false discovery for large data generated by tandem mass Cell. Proteomics. 2009; 8: Full Text Full Text PDF PubMed Scopus Google Scholar). the in which proteins can be through shared matches. In that peptide matches are of been to the of in a and peptide matches are the result of sequence the database that are of the of and to many for the protein with or for the protein with proteins from The of sequence in the G.R. The database and database a of data Res. 2010; PubMed Scopus Google Scholar). one with high are the such as for and the such as and Protein is by identification is a statistical the and be of matches are such a because they are more or the database is because of mass that protein on the mass and may be difficult or to and and and or and or A report from database search results can a or minimal to the identified The is database that one or more of the identified The minimal is the set of database that for the identified The for a minimal is as or the of The is for the of or evidence from be used to which proteins were present in the a is by because of the that a using the data may that the sample a of the proteins and are many to a of proteins that for the observed that a minimal is an in or the of the search results for Aebersold R. The and of peptide-centric 2010; PubMed Scopus Google Scholar). is for the protein with the number of high peptide matches or the sequence to be as by the in a protein score or from the or of the assigned peptide matches is a of a that the proteins with the number of peptide matches to the is protein have In a large scale the minimal protein may or of proteins, by protein score can proteins to such as the of difficult to the of a shotgun is the of a of proteins for a is to into a is of the with of the is to present the search results in the to and the evidence for proteins of The report such which proteins to evidence for a particular of protein a or proteins have been or identified proteins into families on sequence the of a result report because to the proteins of using sequence is is by searching a such as or by using a that database into such as T. E. K. V. D.M. M. S. M. Geer L.Y. W. D. D.J. T. A. V. J. A. E. M. K. D. A. G. T.A. L. W. E. J. Database of the for Res. 2010; PubMed Scopus Google Scholar). Mascot is a data that Basic Local Alignment Search Tool clustering of the proteins found in a database search is a large number of proteins are a is have found that clustering by of shared peptide matches is an for and have by hierarchical clustering of each protein family using the score of the nonshared peptide matches as a distance metric. clustering been to mass in with Peptide Mass R. P. K. C. M. J. and identification of protein using analysis of mass Proteome Res. 2006; PubMed Scopus Google Scholar) to for proteins on shared peptide matches. A new report is in which each protein family is represented by a dendrogram family members are or A greedy algorithm is used to on a minimal of The peptide matches to the proteins in a family can be compared side by side to assess the experimental evidence for each of If the evidence for a family member is considered inadequate, the dendrogram can be cut to reduce the number of distinct family members. of a data set for the of Nesvizhskii A.I. B. Tabb D.L. Kowalak J.A. assessing the quality and of protein on a Scholar) are used to in the from a were digested with with with an for and and then into by The were on an and were generated by members in a of The Mascot set used from were using Mascot used, which and a database of sequences in which the number of and the of each is the as in the Search Mass Mass of and are from The mass may high for a they were set by of search results and were to the observed sequence were as in the The algorithm used to create the protein family which is of Mascot is in The report is generated by a that the Mascot to data from the Mascot result In the new proteins are grouped into a family they have matches to one or more distinct peptide sequences in with the in the because they have an of and be used as evidence to or If proteins have the set of peptide the distance is If they have a of shared and nonshared the distance is the of the score of the nonshared matches in one because these one protein a of the on the shared matches. In each distinct peptide sequence is represented by the match with the of or that distance is and the score distance to protein A into a of protein be the as that to a of A. The distance is are to the of proteins which have peptide matches to the with the one of these matches can be the contains from multiple as An is the peptide sequences in of and L. sequences may in the are the is evidence from the mass spectrometry data to the A for proteomic 2010; 10: PubMed Scopus Google Scholar, and protein 2010; Scholar). a match to the distance the the in which have proteins with peptide matches to the the are the the score is and one match a score of and the a score of one of these matches can be is the as they were matches to the that matches to the with the score to a distance of matches to the with a distance that is the score In the distance be If the matches from and be the distance be or on which one to be to one protein into a of the the that matches to peptides are because the is a from a of and In the from spectra are and be to matches to and one or sequences are represented with a score in the search from a or in a then the distance the proteins be In the of be as match to the create the a distance which is the distance each of The proteins by the distance are to create a and the of the from the is the score distance the The proteins are from the and the the new and proteins The is one the dendrogram is the is to is to the of the and are many to the as to In the of the proteins are in of which be from the of the dendrogram is for members of a family to have shared matches. protein in the family is to by of shared or they have been grouped that are to be shared matches of family members. More a family member to have shared matches with can the proteins that the member into the family are to as such a one family from searching the data the database proteins the report the dendrogram is for each with and for the proteins for each family protein is a of the proteins in in an of the the report to a of proteins, for families of The report is with a of families can be to a number or families in a If a family is of the can be by on the to that of the is from the protein that family contains the family is A of the proteins with additional and for proteins is by a of the peptide matches. the for the been that peptide matches with a are for such are the of the each of the proteins, the protein score and are of the number of matches and the number of distinct sequences. In each the number is the the number in is the for matches the have the report to matches the are the in the proteins are in of protein and can be that family member is represented by matches to distinct sequences compared with family member which A of the of each protein is by Protein T. T. T. J. M. protein for estimation of protein in by the number of peptides protein. Mol. 2005; 4: Google Scholar). of the in the peptide match be are matches to the sequence with the and the match is by on the in the the to the matches to the matches MS/MS are in the result the match assigned to the protein is by on the in the the to the matches The contains a the peptide match is to the each family is grouped on matches the matches can be and such matches have been in the The a the family members and the peptide matches. A the of a and that the peptide match is found in the family the be in a and many can the the can be used to a set of Peptide match are the match is found in one of the proteins, the number of and the to family members and that they have distinct sequences in and that have to matches to distinct sequences to into a of The score for these sequence which to the distance on the dendrogram from the to the which these family members are the dendrogram into a of and reduce the family to members. A to in a of identified proteins to one is to these matches to distinct sequences as the peptide match of these sequences have of the score from with the score of to an of In such a the to the matches to the is because be that match with a score that to a If in using match as evidence for be is the A be to a and a peptide is found in one of the family the that the peptide contains a can be by on In a search of using the match is to the sequence are comparing these are in and is a match on one to as evidence that the sample in the is family of proteins, in the large in protein score family member score and member score In a report by protein these proteins be grouped into a the members are by a score of to a If for with a score of and an of family member represented by distinct a of is a the evidence for the of is and many to In a of proteins, such to be to that can be in how the dendrogram can be used as a to such a score of or or is family members for which is evidence can be to The to the quality of protein inference on a complex data and the results were as a that the of and false proteins by the The a of the results by the with of proteins on shared matches using An the results and the of proteins from the Mascot family report is as a peptide of as by the database are proteins with matches to or more distinct peptide sequences. In the the a score of protein are into families family members. Relative to the to and false in the proteins for the number of false more the of a number of false The how have been assigned to is to which family members be considered the false to how an algorithm approaches an of proteins is to the results for the search of in which the are of the with the the sequences from by a of to and sequences were in then the in were the in If these were to be the results be by the in from the in the number of a match that the in one of the be in the search of a for and the sequences in that were of the the of the database for the sample from to the were in with a to that a of Mascot a search of a the to be to a peptide and of the matches that protein matches to more than one distinct peptide a that a protein matches to distinct peptide that false proteins are the number of proteins as a of the of the The Peptide report been of Mascot the of with nonshared matches with the can one protein from a of the algorithm to The number of proteins from in the database to in the Mascot Peptide report the the the for in are on matches the The Peptide more approaches a minimal of proteins are Peptide matches that are a are and a protein is contains one match that is a match that is the for the and a match to the in protein a of proteins, which with database to The the of the proteins and proteins that be proteins for because on the in which the proteins in the The of are for family members in the new report and for The the peptide and of family members is that are proteins in particular In the peptide to one or is by the proteins that in the search of the is as family member because of a match to is the match to the and are matches for from the Peptide The of families is with database to for the because proteins from as to families that shared matches in proteins from the number of protein for as a of database in from to in The of the by and the of the Peptide with Peptide peptide matches with the and proteins that one match that is members in the new in the new Peptide Peptide in a new clustering by shared peptide matches with clustering by sequence the family report for a search of a database with the results of the search the original into T. E. K. V. D.M. M. S. M. Geer L.Y. W. D. D.J. T. A. V. J. A. E. M. K. D. A. G. T.A. L. W. E. J. Database of the for Res. 2010; PubMed Scopus Google Scholar) is a for into a set of The sequence database the of sequences from sequences. The the for the families in the report the is set to a for peptide matches and are cut a score of If the clustering methods have the members of each family to to the of the families have a to a and to multiple and The is one or shared matches proteins that have of and for the families in the results from searching the data of sequences from protein family family family family family family family family to protein protein protein protein to family A large large large Protein protein in a new The of the matches for family In is assigned to is assigned to The of the sequences using M.A. G. R. P.A. H. A. R. Gibson and 2007; PubMed Scopus Google Scholar) is in the of with the matches in The in which have matches is by a The of the sequences have are peptide matches in these the distance the sequences in the dendrogram is is to have evidence for is one of nonshared peptide and these by a the the for the matches are with of the of are matches to each and the sequences are in which be the one match is the that matches are to the sequence one or are to the in a that a mass of is in that for are multiple are If the to be to more data in a clustering these proteins a to present the the sequences into of families and a on the basis of shared peptide matches proteins that to to the proteins are into family and is to be expected that shared matches be than sequence which to protein sequences. Using shared the of proteins are discrimination can be for proteins with high matches to distinct sequences are for a family to have nonshared match for each member one shared match to families are more the of the is high for one protein in each The clustering and peptide is more If is to for matches to peptide sequences that are as to by in proteins, these create false protein matches and false The peptide in Mascot a of and is difficult to a match to the score that a can is with mass is to the peptide the peptide is high and the number of spectra is large compared with the of the an of an the peptide set to for a search of spectra a family of members and the peptide from to many false the family members. the peptide to the family to be the a of the peptide and the the number of spectra and the number of in the than a peptides of which have of in proteins by false peptide matches can to false proteins, they can false proteins are or more matches to a with for peptide sequences. The that matches with the score are as which is for and false and they a match high the peptide the peptide sequence in proteins, the is the peptide match is the is to a which matches to quality spectra are is for a to matches to sequences with that false peptide matches create false in The is have peptide matches to or more sequences with high shared matches. can the of by for members from families that are assigned to the are such in the the sequences to have which that are shared peptides in the protein. and to the acid sequences have they have a shared peptide in with these on the basis of shared peptide matches. A new report been that to present database search results in a more of peptide match data for protein A greedy set cover algorithm is used to create a minimal set of proteins, grouped into families on the basis of shared peptide matches. families with multiple hierarchical clustering is using the of nonshared peptide matches as a distance metric. how family members are and can be cut to members for which is to be the to proteins of in large data the of proteins may be of