On Using and Extending the MsBackendRawFileReader Backend

Tobias Kockmann1* and Christian Panse1,2**

1Functional Genomics Center Zurich - Swiss Federal Institute of Technology in Zurich
2Swiss Institute of Bioinformatics

*Tobias.Kockmann@fgcz.ethz.ch
**cp@fgcz.ethz.ch

Abstract

MsBackendRawFileReader implements an MsBackend for the Spectra package using Thermo Fisher Scientific’s NewRawFileReader .Net libraries. The package is generalizing the functionallity provided by the rawrr. The vignette utilizes data provided through the tartare package.

Package

MsBackendRawFileReader 1.13.1

The figure below depicts the idea of the Spectra framework. For a detailed description, read (Rainer et al. 2022).

Figure 1: Integration of rawDiag and rawrr into the Spectra ecosystem (by courtesy of Johannes Rainer)

1 Requirements

suppressMessages(
  stopifnot(require(Spectra),
            require(MsBackendRawFileReader),
            require(tartare),
            require(BiocParallel))
)

assemblies aka Common Intermediate Language bytecode The download and install can be done on all platforms using the command: rawrr::installRawFileReaderDLLs()

if (isFALSE(file.exists(rawrr:::.rawrrAssembly()))){
 rawrr::installRawrrExe()
}

2 Load data

# fetch via ExperimentHub
library(ExperimentHub)
eh <- ExperimentHub::ExperimentHub()

query(eh, c('tartare'))

## ExperimentHub with 5 records
## # snapshotDate(): 2024-10-24
## # $dataprovider: Functional Genomics Center Zurich (FGCZ)
## # $species: NA
## # $rdataclass: Spectra
## # additional mcols(): taxonomyid, genome, description,
## #   coordinate_1_based, maintainer, rdatadateadded, preparerclass, tags,
## #   rdatapath, sourceurl, sourcetype 
## # retrieve records with, e.g., 'object[["EH3219"]]' 
## 
##            title                     
##   EH3219 | Q Exactive HF-X mzXML     
##   EH3220 | Q Exactive HF-X raw       
##   EH3221 | Fusion Lumos mzXML        
##   EH3222 | Fusion Lumos raw          
##   EH4547 | Q Exactive HF Orbitrap raw

The RawFileReader libraries require a file extension ending with .raw.

EH3220 <- normalizePath(eh[["EH3220"]])
(rawfileEH3220 <- paste0(EH3220, ".raw"))

## [1] "/home/biocbuild/.cache/R/ExperimentHub/1cd7b350c70a37_3236.raw"

if (!file.exists(rawfileEH3220)){
  file.link(EH3220, rawfileEH3220)
}

EH3222 <- normalizePath(eh[["EH3222"]])
(rawfileEH3222 <- paste0(EH3222, ".raw"))

## [1] "/home/biocbuild/.cache/R/ExperimentHub/18273831991816_3238.raw"

if (!file.exists(rawfileEH3222)){
  file.link(EH3222, rawfileEH3222)
}

EH4547  <- normalizePath(eh[["EH4547"]])
(rawfileEH4547  <- paste0(EH4547 , ".raw"))

## [1] "/home/biocbuild/.cache/R/ExperimentHub/1826043b0ae2f1_4590.raw"

if (!file.exists(rawfileEH4547 )){
  file.link(EH4547 , rawfileEH4547 )
}

3 Usage

Call the constructor using Spectra::backendInitialize, see also (Rainer et al. 2022).

beRaw <- Spectra::backendInitialize(
  MsBackendRawFileReader::MsBackendRawFileReader(),
  files = c(rawfileEH3220, rawfileEH3222, rawfileEH4547))

Call the print method

beRaw

## MsBackendRawFileReader with 32500 spectra
##         msLevel      rtime scanIndex
##       <integer>  <numeric> <integer>
## 1             1 0.00358906         1
## 2             1 0.01189523         2
## 3             1 0.01847028         3
## 4             1 0.02504740         4
## 5             1 0.03161818         5
## ...         ...        ...       ...
## 32496         2    34.9964     21877
## 32497         2    34.9978     21878
## 32498         2    34.9992     21879
## 32499         2    35.0007     21880
## 32500         2    35.0021     21881
##  ... 30 more variables/columns.
## 
## file(s):
## 1cd7b350c70a37_3236.raw
## 18273831991816_3238.raw
## 1826043b0ae2f1_4590.raw

beRaw |> Spectra::spectraVariables()

##  [1] "msLevel"                 "rtime"                  
##  [3] "acquisitionNum"          "scanIndex"              
##  [5] "mz"                      "intensity"              
##  [7] "dataStorage"             "dataOrigin"             
##  [9] "centroided"              "smoothed"               
## [11] "polarity"                "precScanNum"            
## [13] "precursorMz"             "precursorIntensity"     
## [15] "precursorCharge"         "collisionEnergy"        
## [17] "isolationWindowLowerMz"  "isolationWindowTargetMz"
## [19] "isolationWindowUpperMz"  "scanType"               
## [21] "charge"                  "masterScan"             
## [23] "dependencyType"          "monoisotopicMz"         
## [25] "AGC"                     "injectionTime"          
## [27] "resolution"              "isolationWidth"         
## [29] "isolationOffset"         "AGCTarget"              
## [31] "collisionEnergyList"     "AGCFill"                
## [33] "isStepped"

4 Application example

4.1 Peptide Identification

Here we reproduce the Figure 2 of Kockmann and Panse (2021) rawrr. The MsBackendRawFileReader ships with a filterScan method using functionality provided by the C# libraries by Thermo Fisher Scientific Shofstahl (2016).

(S <- (beRaw |>  
   filterScan("FTMS + c NSI Full ms2 487.2567@hcd27.00 [100.0000-1015.0000]") )[437]) |> 
  plotSpectra()

# supposed to be scanIndex 9594
S

## MsBackendRawFileReader with 1 spectra
##     msLevel     rtime scanIndex
##   <integer> <numeric> <integer>
## 1         2   15.4204      9594
##  ... 30 more variables/columns.
## 
## file(s):
## 1826043b0ae2f1_4590.raw

# add yIonSeries to the plot
(yIonSeries <- protViz::fragmentIon("LGGNEQVTR")[[1]]$y[1:8])

## [1] 175.1190 276.1666 375.2350 503.2936 632.3362 746.3791 803.4006 860.4221

names(yIonSeries) <- paste0("y", seq(1, length(yIonSeries)))
abline(v = yIonSeries, col='#DDDDDD88', lwd=5)
axis(3, yIonSeries, names(yIonSeries))

Figure 2: Peptide spectrum match
The vertical grey lines indicate the in-silico computed y-ions of the peptide precusor LGGNEQVTR++ as calculated by the protViz package.

4.2 Class extension

For demonstration reasons, we extent the MsBackend class by a filter method. The filterIons function returns spectra if and only if all fragment ions, given as argument, match. We use protViz::findNN binary search method for determining the nearest mZ peak for each ion. If the mass error between an ion and an mz value is less than the given mass tolerance, an ion is considered a hit.

setGeneric("filterIons", function(object, ...) standardGeneric("filterIons"))

## [1] "filterIons"

setMethod("filterIons", "MsBackend",
  function(object, mZ=numeric(), tol=numeric(), ...) {
    
    keep <- lapply(peaksData(object, BPPARAM = bpparam()),
                   FUN=function(x){
       NN <- protViz::findNN(mZ, x[, 1])
       hit <- (error <- mZ - x[NN, 1]) < tol & x[NN, 2] >= quantile(x[, 2], .9)
       if (sum(hit) == length(mZ))
         TRUE
       else
         FALSE
                   })
    object[unlist(keep)]
  })

The lines below implement a simple targeted peptide search engine. The R code snippet takes as input a MsBackendRawFileReader object containing 32500 spectra and y-fragment-ion mZ values determined for LGGNEQVTR++.

start_time <- Sys.time()
X <- beRaw |> 
  MsBackendRawFileReader::filterScan("FTMS + c NSI Full ms2 487.2567@hcd27.00 [100.0000-1015.0000]") |>
  filterIons(yIonSeries, tol = 0.005) |> 
  Spectra::Spectra() |>
  Spectra::peaksData() 
end_time <- Sys.time()

The defined filterIons method runs on 995 input spectra and returns 4 spectra.

The runtime is shown below.

end_time - start_time

## Time difference of 3.333102 secs

Next, we define and apply a method for graphing LGGNEQVTR peptide spectrum matches. Also, the function returns some statistics of the match.

## A helper plot function to visualize a peptide spectrum match for 
## the LGGNEQVTR peptide.
.plot.LGGNEQVTR <- function(x){
  
  yIonSeries <- protViz::fragmentIon("LGGNEQVTR")[[1]]$y[1:8]
  names(yIonSeries) <- paste0("y", seq(1, length(yIonSeries)))
  
  plot(x, type = 'h', xlim = range(yIonSeries))
  abline(v = yIonSeries, col = '#DDDDDD88', lwd=5)
  axis(3, yIonSeries, names(yIonSeries))
  
  # find nearest mZ value
  idx <- protViz::findNN(yIonSeries, x[,1])
  
  data.frame(
    ion = names(yIonSeries),
    mZ.yIon = yIonSeries,
    mZ = x[idx, 1],
    intensity = x[idx, 2]
  )
}

Figure 3: Visualizing of the LGGNEQVTR spectrum matches

stats::aggregate(mZ ~ ion, data = XC, FUN = base::mean)

##   ion       mZ
## 1  y1 175.1190
## 2  y2 276.1665
## 3  y3 375.2349
## 4  y4 503.2936
## 5  y5 632.3362
## 6  y6 746.3791
## 7  y7 803.4003
## 8  y8 860.4216

stats::aggregate(intensity ~ ion, data = XC, FUN = base::max)

##   ion intensity
## 1  y1   1505214
## 2  y2   2583122
## 3  y3   2364014
## 4  y4   3179124
## 5  y5   2286947
## 6  y6   1236341
## 7  y7   4586484
## 8  y8  12894520

We demonstrate the Spectra::combinePeaks method and aggregate the four spectra into a single peak matrix. The statistics returned by .plot.LGGNEQVTR() should be identical to the aggregation code snippet output above.

X |>
  Spectra::combinePeaks(ppm=10, intensityFun=base::max) |>
  .plot.LGGNEQVTR()

## Error in h(simpleError(msg, call)): error in evaluating the argument 'x' in selecting a method for function 'plot': unable to find an inherited method for function 'combinePeaks' for signature 'object = "SimpleList"'

4.3 Export Mascot Generic Format File

Below we demonstrate the interaction with the MsBackendMgf package while composing a Mascot Generic Format mgf file which is compatible with conducting an MS/MS Ions Search using Mascot Server (>=2.7) Perkins et al. (1999).

if (require(MsBackendMgf)){
    ## Map Spectra variables to Mascot Server compatible vocabulary.
    map <- c(custom = "TITLE",
             msLevel = "CHARGE",
             scanIndex = "SCANS",
             precursorMz = "PEPMASS",
             rtime = "RTINSECONDS")
    
    ## Compose custom TITLE
    beRaw$custom <- paste0("File: ", beRaw$dataOrigin, " ; SpectrumID: ", S$scanIndex)
    
    (mgf <- tempfile(fileext = '.mgf'))
    
    (beRaw |>
            filterScan("FTMS + c NSI Full ms2 487.2567@hcd27.00 [100.0000-1015.0000]") )[437] |>
        Spectra::Spectra() |>
        Spectra::selectSpectraVariables(c("rtime", "precursorMz",
                                          "precursorCharge", "msLevel", "scanIndex", "custom")) |>
        MsBackendMgf::export(backend = MsBackendMgf::MsBackendMgf(),
                             file = mgf, map = map)
    readLines(mgf) |> head(12)
    readLines(mgf) |> tail()
}

## Loading required package: MsBackendMgf

## [1] "862.427612304688 154045.78125" "870.404846191406 159569.8125" 
## [3] "871.395141601562 196302.6875"  "880.671020507812 65916"       
## [5] "END IONS"                      ""

To extract all tandem spectra, you can use the code snippets below

S <- Spectra::backendInitialize(
  MsBackendRawFileReader::MsBackendRawFileReader(),
  files = c(rawfileEH4547)) |>
  Spectra() 

S

## MSn data (Spectra) with 21881 spectra in a MsBackendRawFileReader backend:
##         msLevel      rtime scanIndex
##       <integer>  <numeric> <integer>
## 1             1 0.00258665         1
## 2             2 0.00686231         2
## 3             2 0.00828043         3
## 4             2 0.00971328         4
## 5             2 0.01112509         5
## ...         ...        ...       ...
## 21877         2    34.9964     21877
## 21878         2    34.9978     21878
## 21879         2    34.9992     21879
## 21880         2    35.0007     21880
## 21881         2    35.0021     21881
##  ... 30 more variables/columns.
## 
## file(s):
## 1826043b0ae2f1_4590.raw

S |>
  MsBackendMgf::export(backend = MsBackendMgf::MsBackendMgf(),
                       file = mgf,
                       map = map)

Next, we generate an mgf file for each scan type. This is helpful, e.g., for optimizing search settings tandem mass spectrometry sequence database search tool as comet Eng, Jahan, and Hoopmann (2012) or mascot server Perkins et al. (1999).

## Define scanType patterns
scanTypePattern <- list(
  EThcD.lowres = "ITMS.+sa Full ms2.+@etd.+@hcd.+",
  ETciD.lowres = "ITMS.+sa Full ms2.+@etd.+@cid.+",
  CID.lowres = "ITMS[^@]+@cid[^@]+$",
  HCD.lowres = "ITMS[^@]+@hcd[^@]+$",
  EThcD.highres = "FTMS.+sa Full ms2.+@etd.+@hcd.+",
  HCD.highres = "FTMS[^@]+@hcd[^@]+$"
)

beRaw <- Spectra::backendInitialize(
  MsBackendRawFileReader::MsBackendRawFileReader(),
  files = c(rawrr::sampleFilePath()))

beRaw <- Spectra::backendInitialize(
  MsBackendRawFileReader::MsBackendRawFileReader(),
  files = rawrr::sampleFilePath())

beRaw$custom <- paste0("File: ", gsub("/srv/www/htdocs/Data2San/", "", beRaw$dataOrigin), " ; SpectrumID: ", beRaw$scanIndex)

.generate_mgf <- function(ext, pattern,  dir=tempdir(), ...){
  mgf <- file.path(dir, paste0(sub("\\.raw", "", unique(basename(beRaw$dataOrigin))),
                               ".", ext, ".mgf"))

  idx <- beRaw$scanType |> grepl(patter=pattern)

  if (sum(idx) == 0) return (NULL)

  message(paste0("Extracting ", sum(idx), " ",
                 pattern, " scans\n\t to file ", mgf, " ..."))

  beRaw[which(idx)] |>
    Spectra::Spectra() |>
    Spectra::selectSpectraVariables(c("rtime", "precursorMz",
    "precursorCharge", "msLevel", "scanIndex", "custom")) |>
    MsBackendMgf::export(backend = MsBackendMgf::MsBackendMgf(),
                         file = mgf,
                         map = map)

  mgf
}

#mapply(ext = names(scanTypePattern),
#      scanTypePattern,
#       FUN = .generate_mgf) |>
#  lapply(FUN = function(f){if (file.exists(f)) {readLines(f) |> head()}})

4.4 Procesing queue

Given the task, we want to filter an MS2 of peak list recorded on an Orbitrap device to be interested only in the top peak within 100 Da mass windows. The following code snippet will demonstrate a solution.

## Define a function that takes a matrix as input and derives
## the top n most intense peaks within a mass window.
## Of note, here, we require centroided data. (no profile mode!)
MsBackendRawFileReader:::.top_n

## function (x, n = 10, mass_window = 100, ...) 
## {
##     if (nrow(x) < n) {
##         return(x)
##     }
##     idx <- unlist(lapply(seq(0, 2000, by = mass_window), function(mZ) {
##         i <- which((mZ < x[, 1] & x[, 1] <= mZ + mass_window))
##         r <- i[order(x[, 2][i], decreasing = TRUE)]
##         if (length(x[, 2]) > length(i)) 
##             return(r[1:n])
##         return(r)
##     }, ...))
##     x[sort(idx[!is.na(idx)]), ]
## }
## <bytecode: 0x64c3113bc870>
## <environment: namespace:MsBackendRawFileReader>

We add our custom code to the processing queue of the Spectra object. Of note, we use n = 1 in praxis n = 10 for a 100 Da mass window, which seems to be a practical choice.

S_2 <- Spectra::addProcessing(S, MsBackendRawFileReader:::.top_n, n = 1) 

The plot below displays a visual control of the custom filter function top_n. On the top is the original spectrum, and the filtered one is on the bottom. A point indicates peaks that match.

Spectra::plotSpectraMirror(S[9594], S_2[9594], ppm = 50)

Figure 4: Spectra mirror plot of the filtered (bottom) and unfiltered scan 9594

The following snippet prints the values of the filtered peaklist and the mZ values of the y-ions.

S_2[9594] |> mz() |> unlist()

## [1] 171.1129 276.1667 375.2351 486.2656 503.2942 632.3369 746.3797 860.4223

yIonSeries

##       y1       y2       y3       y4       y5       y6       y7       y8 
## 175.1190 276.1666 375.2350 503.2936 632.3362 746.3791 803.4006 860.4221

5 Evaluation

5.1 Efficiency - I/O Benchmark

When reading spectra the MsBackendRawFileReader:::.RawFileReader_read_peaks method is calling the rawrr::readSpectrum method.

The figure below displays the time performance for reading a single spectrum in dependency from the chunk size (how many spectra are read in one function call) for reading different numbers of overall spectra.

ioBm <- file.path(system.file(package = 'MsBackendRawFileReader'),
               'extdata', 'specs.csv') |>
  read.csv2(header=TRUE)

# perform and include a local IO benchmark
ioBmLocal <- ioBenchmark(1000, c(32, 64, 128, 256), rawfile = rawfileEH4547)


lattice::xyplot((1 / as.numeric(time)) * workers ~ size | factor(n) ,
                group = host,
                data = rbind(ioBm, ioBmLocal),
                horizontal = FALSE,
        scales=list(y = list(log = 10)),
                auto.key = TRUE,
                layout = c(3, 1),
                ylab = 'spectra read in one second',
                xlab = 'number of spectra / file')

Figure 5: I/O Benchmark
The XY plot graphs how many spectra the backend can read in one second versus the chunk size of the rawrr::readSpectrum method for different compute architectures.

5.2 Effectiveness

We compare the output of the Thermo Fischer Scientific raw files versus their corresponding mzXML files using Spectra::MsBackendMzR relying on the mzR package.

mzXMLEH3219 <- normalizePath(eh[["EH3219"]])

## see ?tartare and browseVignettes('tartare') for documentation

## loading from cache

mzXMLEH3221 <- normalizePath(eh[["EH3221"]])

## see ?tartare and browseVignettes('tartare') for documentation
## loading from cache

if (require(mzR)){
  beMzXML <- Spectra::backendInitialize(
    Spectra::MsBackendMzR(),
    files = c(mzXMLEH3219))
  
  beRaw <- Spectra::backendInitialize(
    MsBackendRawFileReader::MsBackendRawFileReader(),
    files = c(rawfileEH3220))
  
  intensity.xml <- sapply(intensity(beMzXML[1:100]), sum)
  intensity.raw <- sapply(intensity(beRaw[1:100]), sum)
  
  plot(intensity.xml ~ intensity.raw, log = 'xy', asp = 1,
    pch = 16, col = rgb(0.5, 0.5, 0.5, alpha=0.5), cex=2)
  abline(lm(intensity.xml ~ intensity.raw), 
    col='red')
}

Figure 6: Aggregated intensities mzXML versus raw of the 1st 100 spectra

Are all scans of the raw file in the mzXML file?

if (require(mzR)){
  table(scanIndex(beRaw) %in% scanIndex(beMzXML))
}

## 
## FALSE  TRUE 
##   113  1764

Session information

sessionInfo()

## R Under development (unstable) (2024-10-21 r87258)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.1 LTS
## 
## Matrix products: default
## BLAS:   /home/biocbuild/bbs-3.21-bioc/R/lib/libRblas.so 
## LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.12.0
## 
## 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       
## 
## time zone: America/New_York
## tzcode source: system (glibc)
## 
## attached base packages:
## [1] stats4    stats     graphics  grDevices utils     datasets  methods  
## [8] base     
## 
## other attached packages:
##  [1] mzR_2.41.0                    Rcpp_1.0.13-1                
##  [3] MsBackendMgf_1.15.0           tartare_1.21.0               
##  [5] ExperimentHub_2.15.0          AnnotationHub_3.15.0         
##  [7] BiocFileCache_2.15.0          dbplyr_2.5.0                 
##  [9] MsBackendRawFileReader_1.13.1 Spectra_1.17.0               
## [11] BiocParallel_1.41.0           S4Vectors_0.45.0             
## [13] BiocGenerics_0.53.1           generics_0.1.3               
## [15] BiocStyle_2.35.0             
## 
## loaded via a namespace (and not attached):
##  [1] tidyselect_1.2.1        dplyr_1.1.4             blob_1.2.4             
##  [4] filelock_1.0.3          Biostrings_2.75.1       fastmap_1.2.0          
##  [7] digest_0.6.37           mime_0.12               lifecycle_1.0.4        
## [10] cluster_2.1.6           ProtGenerics_1.39.0     KEGGREST_1.47.0        
## [13] RSQLite_2.3.7           magrittr_2.0.3          compiler_4.5.0         
## [16] rlang_1.1.4             sass_0.4.9              tools_4.5.0            
## [19] utf8_1.2.4              yaml_2.3.10             knitr_1.48             
## [22] bit_4.5.0               curl_6.0.0              withr_3.0.2            
## [25] purrr_1.0.2             grid_4.5.0              fansi_1.0.6            
## [28] MASS_7.3-61             tinytex_0.54            cli_3.6.3              
## [31] rmarkdown_2.29          crayon_1.5.3            httr_1.4.7             
## [34] rawrr_1.15.6            ncdf4_1.23              DBI_1.2.3              
## [37] cachem_1.1.0            zlibbioc_1.53.0         parallel_4.5.0         
## [40] AnnotationDbi_1.69.0    BiocManager_1.30.25     XVector_0.47.0         
## [43] vctrs_0.6.5             jsonlite_1.8.9          bookdown_0.41          
## [46] IRanges_2.41.0          bit64_4.5.2             clue_0.3-65            
## [49] magick_2.8.5            protViz_0.7.9           jquerylib_0.1.4        
## [52] glue_1.8.0              codetools_0.2-20        BiocVersion_3.21.1     
## [55] GenomeInfoDb_1.43.0     UCSC.utils_1.3.0        tibble_3.2.1           
## [58] pillar_1.9.0            rappdirs_0.3.3          htmltools_0.5.8.1      
## [61] GenomeInfoDbData_1.2.13 R6_2.5.1                evaluate_1.0.1         
## [64] Biobase_2.67.0          lattice_0.22-6          highr_0.11             
## [67] png_0.1-8               memoise_2.0.1           bslib_0.8.0            
## [70] MetaboCoreUtils_1.15.0  xfun_0.49               MsCoreUtils_1.19.0     
## [73] fs_1.6.5                pkgconfig_2.0.3

References

Eng, Jimmy K., Tahmina A. Jahan, and Michael R. Hoopmann. 2012. “Comet: An Open-Source MS/MS Sequence Database Search Tool.” PROTEOMICS 13 (1): 22–24. https://doi.org/10.1002/pmic.201200439.

Kockmann, Tobias, and Christian Panse. 2021. “The Rawrr R Package: Direct Access to Orbitrap Data and Beyond.” Journal of Proteome Research. https://doi.org/10.1021/acs.jproteome.0c00866.

Perkins, David N., Darryl J. C. Pappin, David M. Creasy, and John S. Cottrell. 1999. “Probability-Based Protein Identification by Searching Sequence Databases Using Mass Spectrometry Data.” Electrophoresis 20 (18): 3551–67. https://doi.org/10.1002/(sici)1522-2683(19991201)20:18<3551::aid-elps3551>3.0.co;2-2.

Rainer, Johannes, Andrea Vicini, Liesa Salzer, Jan Stanstrup, Josep M. Badia, Steffen Neumann, Michael A. Stravs, et al. 2022. “A Modular and Expandable Ecosystem for Metabolomics Data Annotation in r.” Metabolites 12: 173. https://doi.org/10.3390/metabo12020173.

Shofstahl, Jim. 2016. “New RawFileReader from Thermo Fisher Scientific.” 2016. https://planetorbitrap.com/rawfilereader.