Package: PSMatch
Authors: Laurent Gatto [aut, cre] (https://orcid.org/0000-0002-1520-2268), Johannes Rainer [aut] (https://orcid.org/0000-0002-6977-7147), Sebastian Gibb [aut] (https://orcid.org/0000-0001-7406-4443), Samuel Wieczorek [ctb], Thomas Burger [ctb]
Last modified: 2022-11-01 15:21:05
Compiled: Tue Nov 1 18:35:15 2022

1 Installation instructions

To install the package from Bioconductor, make sure you have the BiocManager package, available from CRAN, and then run

BiocManager::install("PSMatch")

2 Introduction

This vignette is one among several illustrating how to use the PSMatch package, focusing on the handling and processing of proteomics identification data using the PSM class. For a general overview of the package, see the PSMatch package manual page (?PSMatch) and references therein.

3 Handling and processing identification data

3.1 Loading PSM data

We are going to use an mzid file from the msdata package.

f <- msdata::ident(full.names = TRUE, pattern = "TMT")
basename(f)
## [1] "TMT_Erwinia_1uLSike_Top10HCD_isol2_45stepped_60min_01-20141210.mzid"

The PSM() function parses one or multiple mzid files and returns an object of class PSM. This class is a simple extension of the DFrame class, representing the peptide-spectrum matches in a tabular format.

library("PSMatch")
id <- PSM(f)
id
## PSM with 5802 rows and 35 columns.
## names(35): sequence spectrumID ... subReplacementResidue subLocation

This table contains 5802 matches for 5343 scans and 4938 peptides sequences, each annotated by 35 variables.

nrow(id) ## number of matches
## [1] 5802
length(unique(id$spectrumID)) ## number of scans
## [1] 5343
length(unique(id$sequence))   ## number of peptide sequences
## [1] 4938
names(id)
##  [1] "sequence"                 "spectrumID"              
##  [3] "chargeState"              "rank"                    
##  [5] "passThreshold"            "experimentalMassToCharge"
##  [7] "calculatedMassToCharge"   "peptideRef"              
##  [9] "modNum"                   "isDecoy"                 
## [11] "post"                     "pre"                     
## [13] "start"                    "end"                     
## [15] "DatabaseAccess"           "DBseqLength"             
## [17] "DatabaseSeq"              "DatabaseDescription"     
## [19] "scan.number.s."           "acquisitionNum"          
## [21] "spectrumFile"             "idFile"                  
## [23] "MS.GF.RawScore"           "MS.GF.DeNovoScore"       
## [25] "MS.GF.SpecEValue"         "MS.GF.EValue"            
## [27] "MS.GF.QValue"             "MS.GF.PepQValue"         
## [29] "modPeptideRef"            "modName"                 
## [31] "modMass"                  "modLocation"             
## [33] "subOriginalResidue"       "subReplacementResidue"   
## [35] "subLocation"

The PSM data are read as is, without any filtering. As we can see below, we still have all the hits from the forward and reverse (decoy) databases.

table(id$isDecoy)
## 
## FALSE  TRUE 
##  2906  2896

3.2 Keeping all matches

The data also contains multiple matches for several spectra. The table below shows the number of individual MS scans that have 1, 2, … up to 5 matches.

table(table(id$spectrumID))
## 
##    1    2    3    4    5 
## 4936  369   26   10    2

More specifically, we can see below how scan 1774 has 4 matches, all to sequence RTRYQAEVR, which itself matches to 4 different proteins:

i <- grep("scan=1774", id$spectrumID)
id[i, ]
## PSM with 4 rows and 35 columns.
## names(35): sequence spectrumID ... subReplacementResidue subLocation
id[i, "DatabaseAccess"]
## [1] "ECA2104" "ECA2867" "ECA3427" "ECA4142"

If the goal is to keep all the matches, but arranged by scan/spectrum, one can reduce the DataFrame object by the spectrumID variable, so that each scan correponds to a single row that still stores all values1 The rownames aren’t needed here and are removed to reduce to output in the the next code chunks displaying parts of id2.:

id2 <- reducePSMs(id, id$spectrumID)
rownames(id2) <- NULL ## rownames not needed here
dim(id2)
## [1] 5343   35

The resulting object contains a single entrie for scan 1774 with information for the multiple matches stored as a list within the table cell.

j <- grep("scan=1774", id2$spectrumID)
id2[j, ]
## Reduced PSM with 1 rows and 35 columns.
## names(35): sequence spectrumID ... subReplacementResidue subLocation
id2[j, "DatabaseAccess"]
## CharacterList of length 1
## [["controllerType=0 controllerNumber=1 scan=1774"]] ECA2104 ECA2867 ECA3427 ECA4142

The identification data could be used to annotate an raw mass spectrometry Spectra object (see the Spectra::joinSpectraData() function for details).

3.3 Filtering data

Often, the PSM data is filtered to only retain reliable matches. The MSnID package can be used to set thresholds to attain user-defined PSM, peptide or protein-level FDRs. Here, we will simply filter out wrong or the least reliable identifications.

3.3.1 Remove decoy hits

id <- filterPsmDecoy(id)
## Removed 2896 decoy hits.
id
## PSM with 2906 rows and 35 columns.
## names(35): sequence spectrumID ... subReplacementResidue subLocation

3.3.2 Keep first rank matches

id <- filterPsmRank(id)
## Removed 155 PSMs with rank > 1.
id
## PSM with 2751 rows and 35 columns.
## names(35): sequence spectrumID ... subReplacementResidue subLocation

3.3.3 Remove shared peptides

The data still contains shared peptides, i.e. those that different proteins. For example QKTRCATRAFKANKGRAR matches proteins ECA2869 and ECA4278.

id[id$sequence == "QKTRCATRAFKANKGRAR", "DatabaseAccess"]
## [1] "ECA2869" "ECA4278"

We can filter these out to retain unique peptides.

id <- filterPsmShared(id)
## Removed 85 shared peptides.
id
## PSM with 2666 rows and 35 columns.
## names(35): sequence spectrumID ... subReplacementResidue subLocation

This last filtering leaves us with 2666 PSMs.

Note that the ConnectedComponents approach defined in this package allows one to explore protein groups defined by such shared peptides more thoroughly and make informed decision as to which shared peptides to retain and which ones to drop. For details see the related vignette or the ConnectedComponents manual page.

3.3.4 All filters in one function

This can also be achieved with the filterPSMs() function:

id <- PSM(f)
filterPSMs(id)
## Starting with 5802 PSMs:
## Removed 2896 decoy hits.
## Removed 155 PSMs with rank > 1.
## Removed 85 shared peptides.
## 2666 PSMs left.
## PSM with 2666 rows and 35 columns.
## names(35): sequence spectrumID ... subReplacementResidue subLocation

4 The mzR and mzID parsers

The PSM() function can take two different values for the parser parameter, namely "mzR" (the default value) and "mzID".

  • mzR uses the openIDfile() function from the mzR to parse the mzId file(s), and then coerces the data to a data.frame which is eventually returned as a PSM object. The parser function uses dedicated code from the Proteowizard project (included in mzR) and is generally the fastest approach.

  • mzID parses the mzId file with mzID() function from the mzID package, and then flattens the data to a data.frame with mzID::flatten() and eventuelly returns a PSM object. The mzID package relies on the XML package. Is is slower but is is more robust to variations in the mzID implementation, as is thus a useful backup when the mzR backend fails.

system.time(id1 <- PSM(f, parser = "mzR"))
##    user  system elapsed 
##   0.567   0.040   0.607
system.time(id2 <- PSM(f, parser = "mzID"))
## Loading required namespace: mzID
## reading TMT_Erwinia_1uLSike_Top10HCD_isol2_45stepped_60min_01-20141210.mzid... DONE!
##    user  system elapsed 
##  10.861   0.440  11.302

Other differences in the two parsers include the columns that are returned, the way they name them, and, as will shown below the matches that are returned. Note for instance (and this will be important later), that there is no equivalent of "modLocation" in id2.

names(id1)
##  [1] "sequence"                 "spectrumID"              
##  [3] "chargeState"              "rank"                    
##  [5] "passThreshold"            "experimentalMassToCharge"
##  [7] "calculatedMassToCharge"   "peptideRef"              
##  [9] "modNum"                   "isDecoy"                 
## [11] "post"                     "pre"                     
## [13] "start"                    "end"                     
## [15] "DatabaseAccess"           "DBseqLength"             
## [17] "DatabaseSeq"              "DatabaseDescription"     
## [19] "scan.number.s."           "acquisitionNum"          
## [21] "spectrumFile"             "idFile"                  
## [23] "MS.GF.RawScore"           "MS.GF.DeNovoScore"       
## [25] "MS.GF.SpecEValue"         "MS.GF.EValue"            
## [27] "MS.GF.QValue"             "MS.GF.PepQValue"         
## [29] "modPeptideRef"            "modName"                 
## [31] "modMass"                  "modLocation"             
## [33] "subOriginalResidue"       "subReplacementResidue"   
## [35] "subLocation"
names(id2)
##  [1] "spectrumid"                "scan number(s)"           
##  [3] "acquisitionnum"            "passthreshold"            
##  [5] "rank"                      "calculatedmasstocharge"   
##  [7] "experimentalmasstocharge"  "chargestate"              
##  [9] "ms-gf:denovoscore"         "ms-gf:evalue"             
## [11] "ms-gf:pepqvalue"           "ms-gf:qvalue"             
## [13] "ms-gf:rawscore"            "ms-gf:specevalue"         
## [15] "assumeddissociationmethod" "isotopeerror"             
## [17] "isdecoy"                   "post"                     
## [19] "pre"                       "end"                      
## [21] "start"                     "accession"                
## [23] "length"                    "description"              
## [25] "pepseq"                    "modified"                 
## [27] "modification"              "idFile"                   
## [29] "spectrumFile"              "databaseFile"

We also have different number of matches in the two tables:

nrow(id1)
## [1] 5802
nrow(id2)
## [1] 5759
table(id1$isDecoy)
## 
## FALSE  TRUE 
##  2906  2896
table(id2$isdecoy)
## 
## FALSE  TRUE 
##  2886  2873

Let’s first filter the PSM tables to facilitate focus the comparison of relevant scans. Note that the default filterPSMs() arguments are set to work with both parser.

id1_filtered <- filterPSMs(id1)
## Starting with 5802 PSMs:
## Removed 2896 decoy hits.
## Removed 155 PSMs with rank > 1.
## Removed 85 shared peptides.
## 2666 PSMs left.
id2_filtered <- filterPSMs(id2)
## Starting with 5759 PSMs:
## Removed 2873 decoy hits.
## Removed 155 PSMs with rank > 1.
## Removed 85 shared peptides.
## 2646 PSMs left.

As can be seen, we are also left with 2666 vs 2646 PSMs after filtering.

The difference doesn’t stem from different scans, given that the spectum identifiers are identical in both tables:

identical(sort(unique(id1_filtered$spectrumID)),
          sort(unique(id2_filtered$spectrumid)))
## [1] TRUE

The difference is obvious when we tally a table of spectrum id occurences in the filtered tables. In id2_filtered, each scan is unique, i.e matched only once.

anyDuplicated(id2_filtered$spectrumid)
## [1] 0

However, for id1_filtered, we see that some scans are still repeat up to 4 times in the table:

table(table(id1_filtered$spectrumID))
## 
##    1    2    3    4 
## 2630   13    2    1

The example below shows that these differences stem from the modification location ("modLocation"), that is not report by the mzID parser:

k <- names(which(table(id1_filtered$spectrumID) == 4))
id1_filtered[id1_filtered$spectrumID == k, "sequence"]
## [1] "KCNQCLKVACTLFYCK" "KCNQCLKVACTLFYCK" "KCNQCLKVACTLFYCK" "KCNQCLKVACTLFYCK"
id1_filtered[id1_filtered$spectrumID == k, "modLocation"]
## [1]  2  5 10 15
id1_filtered[id1_filtered$spectrumID == k, "modName"]
## [1] "Carbamidomethyl" "Carbamidomethyl" "Carbamidomethyl" "Carbamidomethyl"

If we remove the "modLocation" column, we recoved the same number of PSMs than with the mzID parser.

id1_filtered$modLocation <- NULL
nrow(unique(id1_filtered))
## [1] 2646
nrow(unique(id2_filtered))
## [1] 2646

5 Session information

sessionInfo()
## R version 4.2.1 (2022-06-23)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.5 LTS
## 
## Matrix products: default
## BLAS:   /home/biocbuild/bbs-3.16-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.16-bioc/R/lib/libRlapack.so
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_GB              LC_COLLATE=C              
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] stats4    stats     graphics  grDevices utils     datasets  methods  
## [8] base     
## 
## other attached packages:
##  [1] Spectra_1.8.0               ProtGenerics_1.30.0        
##  [3] BiocParallel_1.32.0         factoextra_1.0.7           
##  [5] ggplot2_3.3.6               QFeatures_1.8.0            
##  [7] MultiAssayExperiment_1.24.0 SummarizedExperiment_1.28.0
##  [9] Biobase_2.58.0              GenomicRanges_1.50.0       
## [11] GenomeInfoDb_1.34.0         IRanges_2.32.0             
## [13] MatrixGenerics_1.10.0       matrixStats_0.62.0         
## [15] PSMatch_1.2.0               S4Vectors_0.36.0           
## [17] BiocGenerics_0.44.0         BiocStyle_2.26.0           
## 
## loaded via a namespace (and not attached):
##  [1] fs_1.5.2                bitops_1.0-7            doParallel_1.0.17      
##  [4] tools_4.2.1             backports_1.4.1         bslib_0.4.0            
##  [7] utf8_1.2.2              R6_2.5.1                DBI_1.1.3              
## [10] lazyeval_0.2.2          colorspace_2.0-3        withr_2.5.0            
## [13] tidyselect_1.2.0        compiler_4.2.1          cli_3.4.1              
## [16] DelayedArray_0.24.0     labeling_0.4.2          bookdown_0.29          
## [19] sass_0.4.2              scales_1.2.1            stringr_1.4.1          
## [22] digest_0.6.30           rmarkdown_2.17          XVector_0.38.0         
## [25] pkgconfig_2.0.3         htmltools_0.5.3         fastmap_1.1.0          
## [28] highr_0.9               rlang_1.0.6             farver_2.1.1           
## [31] jquerylib_0.1.4         generics_0.1.3          jsonlite_1.8.3         
## [34] mzID_1.36.0             car_3.1-1               dplyr_1.0.10           
## [37] RCurl_1.98-1.9          magrittr_2.0.3          GenomeInfoDbData_1.2.9 
## [40] Matrix_1.5-1            Rcpp_1.0.9              munsell_0.5.0          
## [43] fansi_1.0.3             abind_1.4-5             MsCoreUtils_1.10.0     
## [46] lifecycle_1.0.3         stringi_1.7.8           yaml_2.3.6             
## [49] carData_3.0-5           MASS_7.3-58.1           zlibbioc_1.44.0        
## [52] plyr_1.8.7              grid_4.2.1              parallel_4.2.1         
## [55] ggrepel_0.9.1           lattice_0.20-45         mzR_2.32.0             
## [58] magick_2.7.3            knitr_1.40              pillar_1.8.1           
## [61] igraph_1.3.5            ggpubr_0.4.0            ggsignif_0.6.4         
## [64] codetools_0.2-18        XML_3.99-0.12           glue_1.6.2             
## [67] evaluate_0.17           BiocManager_1.30.19     foreach_1.5.2          
## [70] vctrs_0.5.0             gtable_0.3.1            purrr_0.3.5            
## [73] tidyr_1.2.1             clue_0.3-62             assertthat_0.2.1       
## [76] cachem_1.0.6            xfun_0.34               broom_1.0.1            
## [79] AnnotationFilter_1.22.0 msdata_0.37.0           rstatix_0.7.0          
## [82] ncdf4_1.19              tibble_3.1.8            iterators_1.0.14       
## [85] cluster_2.1.4