# MSnbase: MS data processing, visualisation and quantification In MSnbase: Base Functions and Classes for Mass Spectrometry and Proteomics

BiocStyle::markdown()

suppressPackageStartupMessages(library("ggplot2"))
suppressPackageStartupMessages(library("MSnbase"))
suppressPackageStartupMessages(library("zoo"))
suppressPackageStartupMessages(require("Rdisop"))
suppressPackageStartupMessages(require("pRolocdata"))
suppressPackageStartupMessages(require("pRoloc"))
suppressPackageStartupMessages(require("msdata"))
library("grid")
suppressPackageStartupMessages(library("BiocParallel"))

cat(readLines("./Foreword.md"), sep = "\n")

cat(readLines("./Bugs.md"), sep = "\n")


# Introduction {#sec:intro}

r Biocpkg("MSnbase") [@Gatto2012] aims are providing a reproducible research framework to proteomics data analysis. It should allow researcher to easily mine mass spectrometry data, explore the data and its statistical properties and visually display these.

r Biocpkg("MSnbase") also aims at being compatible with the infrastructure implemented in Bioconductor, in particular r Biocpkg("Biobase"). As such, classes developed specifically for proteomics mass spectrometry data are based on the eSet and ExpressionSet classes. The main goal is to assure seamless compatibility with existing meta data structure, accessor methods and normalisation techniques.

This vignette illustrates r Biocpkg("MSnbase") utility using a dummy data sets provided with the package without describing the underlying data structures. More details can be found in the package, classes, method and function documentations. A description of the classes is provided in the MSnbase-development vignette^[in R, open it with vignette("MSnbase-development") or read it online here].

## Speed and memory requirements

Raw mass spectrometry file are generally several hundreds of MB large and most of this is used for binary raw spectrum data. As such, data containers can easily grow very large and thus require large amounts of RAM. This requirement is being tackled by avoiding to load the raw data into memory and using on-disk random access to the content of mzXML/mzML data files on demand. When focusing on reporter ion quantitation, a direct solution for this is to trim the spectra using the trimMz method to select the area of interest and thus substantially reduce the size of the Spectrum objects. This is illustrated in section \@ref(sec:trim).

Parallel processing The independent handling of spectra is ideally suited for parallel processing. The quantify method for example performs reporter peaks quantitation in parallel.

Parallel support is provided by the r Biocpkg("BiocParallel") and various backends including multicore (forking, default on Linux), simple networf network of workstations (SNOW, default on Windows) using sockets, forking or MPI among others. We refer readers to the documentation in r Biocpkg("BiocParallel"). Automatic parallel processing of spectra is only established for a certain number of spectra (per file). This value (default is 1000) can be set with the setMSnbaseParallelThresh function.

In sock-based parallel processing, the main worker process has to start new R instances and connect to them via sock. Sometimes these connections can not be established and the processes get stuck. To test this, users can disable parallel processing by disabling parallel processing with register(SerialParam()). To avoid these deadlocks, it is possible to initiate the parallel processing setup explicitly at the beginning of the script using, for example

library("doParallel")
registerDoParallel(3) ## using 3 slave nodes
register(DoparParam(), default = TRUE)

## rest of script comes below


On-disk access Developmenets in version 2 of the package have solved the memory issue by implementing and on-disk version the of data class storing raw data (MSnExp, see section \@ref(sec:msnexp)), where the spectra a accessed on-disk only when required. The benchmarking vignette compares the on-disk and in-memory implemenatations^[in R, open it with vignette("benchmarking") or read it online here]. See details below.

# Data structure and content {#sec:data}

## Importing experiments {#sec:io}

r Biocpkg("MSnbase") is able to import raw MS data stored in one of the XML-based formats as well as peak lists in the mfg format^[Mascot Generic Format, see http://www.matrixscience.com/help/data_file_help.html#GEN].

Raw data The XML-based formats, mzXML [@Pedrioli2004], mzData [@Orchard2007] and mzML [@Martens2010] can be imported with the readMSData function, as illustrated below (see ?readMSData for more details). To make use of the new on-disk implementation, set mode = "onDisk" in readMSData rather than using the default mode = "inMemory".

file <- dir(system.file(package = "MSnbase", dir = "extdata"),
full.names = TRUE, pattern = "mzXML$") rawdata <- readMSData(file, msLevel = 2, verbose = FALSE)  Only spectra of a given MS level can be loaded at a time by setting the msLevel parameter accordingly in readMSData and in-memory data. In this document, we will use the itraqdata data set, provided with r Biocpkg("MSnbase"). It includes feature metadata, accessible with the fData accessor. The metadata includes identification data for the r length(itraqdata) MS2 spectra. Version 2.0 and later of r Biocpkg("MSnbase") provide a new on-disk data storage model (see the benchmarking vignette for more details). The new data backend is compatible with the orignal in-memory model. To make use of the new infrastructure, read your raw data by setting the mode argument to "onDisk" (the default is still "inMemory" but is likely to change in the future). The new on-disk implementation supports several MS levels in a single raw data object. All existing operations work irrespective of the backend. Peak lists can often be exported after spectrum processing from vendor-specific software and are also used as input to search engines. Peak lists in mgf format can be imported with the function readMgfData (see ?readMgfData for details) to create experiment objects. Experiments or individual spectra can be exported to an mgf file with the writeMgfData methods (see ?writeMgfData for details and examples). Experiments with multiple runs Although it is possible to load and process multiple files serially and later merge the resulting quantitation data as show in section \@ref(sec:combine), it is also feasible to load several raw data files at once. Here, we report the analysis of an LC-MSMS experiment were 14 liquid chromatography (LC) fractions were loaded in memory using readMSData on a 32-cores servers with 128 Gb of RAM. It took about 90 minutes to read the 14 uncentroided mzXML raw files (4.9 Gb on disk in total) and create a 3.3 Gb raw data object (an MSnExp instance, see next section). Quantitation of 9 reporter ions (iTRAQ9 object, see \@ref(sec:reporterions)) for 88690 features was performed in parallel on 16 processors and took 76 minutes. The resulting quantitation data was only 22.1 Mb and could easily be further processed. These number are based on the older in-memory implementation. As shown in the benchmarking vignette, using on-disk data greatly reduces memory requirement and computation time. See also section \@ref(sec:io2) to import quantitative data stored in spreadsheets into R for further processing using r Biocpkg("MSnbase"). The MSnbase-iovignette[in R, open it with vignette("MSnbase-io") or read it online here] gives a general overview of r Biocpkg("MSnbase")'s input/ouput capabilites. See section \@ref(sec:io3) for importing chromatographic data of SRM/MRM experiments. ## Exporting experiments/MS data MSnbase supports also to write MSnExp or OnDiskMSnExp objects to mzML or mzXML files using the writeMSData function. This is specifically useful in workflows in which the MS data was heavily manipulated. Presently, each sample/file is exported into one file. Below we write the data in mzML format to a temporary file. By setting the optional parameter copy = TRUE general metadata (such as instrument info or all data processing descriptions) are copied over from the originating file. writeMSData(rawdata, file = paste0(tempfile(), ".mzML"), copy = TRUE)  ## MS experiments {#sec:msnexp} Raw data is contained in MSnExp objects, that stores all the spectra of an experiment, as defined by one or multiple raw data files. library("MSnbase") itraqdata head(fData(itraqdata))  sz <- sum(sapply(assayData(itraqdata), object.size)) + object.size(itraqdata) sz <- as.numeric(sz) sz <- round(sz/(1024^2), 2)  As illustrated above, showing the experiment textually displays it's content: • Information about the raw data, i.e. the spectra. • Specific information about the experiment processing^[This part will be automatically updated when the object is modified with it's ad hoc methods, as illustrated later.] and package version. This slot can be accessed with the processingData method. • Other meta data, including experimental phenotype, file name(s) used to import the data, protocol data, information about features (individual spectra here) and experiment data. Most of these are implemented as in the eSet class and are described in more details in their respective manual pages. See ?MSnExp and references therein for additional background information. The experiment meta data associated with an MSnExp experiment is of class MIAPE. It stores general information about the experiment as well as MIAPE (Minimum Information About a Proteomics Experiment) information [@Taylor2007, @Taylor2008]. This meta-data can be accessed with the experimentData method. When available, a summary of MIAPE-MS data can be printed with the msInfo method. See ?MIAPE for more details. ## Spectra objects {#sec:spectra} The raw data is composed of the r length(itraqdata) MS spectra. The spectra are named individually (r paste(paste(head(featureNames(itraqdata)),collapse=", "),", ...",sep="")) and stored in a environment. They can be accessed individually with itraqdata[["X1"]] or itraqdata[[1]], or as a list with spectra(itraqdata). As we have loaded our experiment specifying msLevel=2, the spectra will all be of level 2 (or higher, if available). sp <- itraqdata[["X1"]] sp  Attributes of individual spectra or of all spectra of an experiment can be accessed with their respective methods: precursorCharge for the precursor charge, rtime for the retention time, mz for the MZ values, intensity for the intensities, ... see the Spectrum, Spectrum1 and Spectrum2 manuals for more details. peaksCount(sp) head(peaksCount(itraqdata)) rtime(sp) head(rtime(itraqdata))  ## Reporter ions {#sec:reporterions} Reporter ions are defined with the ReporterIons class. Specific peaks of interest are defined by a MZ value, a with around the expected MZ and a name (and optionally a colour for plotting, see section \@ref(sec:plotting)). ReporterIons instances are required to quantify reporter peaks in MSnExp experiments. Instances for the most commonly used isobaric tags like iTRAQ 4-plex and 8-plex and TMT 6- and 10-plex tags are already defined in r Biocpkg("MSnbase"). See ?ReporterIons for details about how to generate new ReporterIons objects. iTRAQ4 TMT10  ## Chromatogram objects {#sec:chromatograms} Chromatographic data, i.e. intensity values along the retention time dimension for a given$m/z$range/slice, can be extracted with the chromatogram method. Below we read a file from the msdata package and extract the (MS level 1) chromatogram. Without providing an$m/z$and a retention time range the function returns the total ion chromatogram (TIC) for each file within the MSnExp or OnDiskMSnExp object. See also section \@ref(sec:io3) for importing chromatographic data from SRM/MRM experiments. f <- c(system.file("microtofq/MM14.mzML", package = "msdata")) mtof <- readMSData(f, mode = "onDisk") mtof_tic <- chromatogram(mtof) mtof_tic  Chromatographic data, represented by the intensity-retention time duplets, is stored in the Chromatogram object. The chromatogram method returns a Chromatograms object (note the s) which holds multiple Chromatogram objects and arranges them in a two-dimensional grid with columns representing files/samples of the MSnExp or OnDiskMSnExp object and rows$m/z$-retention time ranges. In the example above the Chromatograms object contains only a single Chromatogram object. Below we access this chromatogram object. Similar to the Spectrum objects, Chromatogram objects provide the accessor functions intensity and rtime to access the data, as well as the mz function, that returns the$m/z$range of the chromatogram. mtof_tic[1, 1] head(intensity(mtof_tic[1, 1])) head(rtime(mtof_tic[1, 1])) mz(mtof_tic[1, 1])  To extract the base peak chromatogram (the largest peak along the$m/z$dimension for each retention time/spectrum) we set the aggregationFun argument to "max". mtof_bpc <- chromatogram(mtof, aggregationFun = "max")  See the Chromatogram help page and the vignettes from the r Biocpkg("xcms") package for more details and use cases, also on how to extract chromatograms for specific ions. # Plotting raw data {#sec:plotting} ## MS data space {#sec:msmaps} The MSmap class can be used to isolate specific slices of interest from a complete MS acquisition by specifying$m/z$and retention time ranges. One needs a raw data file in a format supported by r Biocpkg("mzR")'s openMSfile (mzML, mzXML, ...). Below we first download a raw data file from the PRIDE repository and create\footnote{This code chunk is not evaluated to avoid repeated downloaded of the raw data file. The M map is provided with the package and loaded to evaluate subsequent code chunks.} an MSmap containing all the MS1 spectra between acquired between 30 and 35 minutes and peaks between 521 and 523$m/z$. See ?MSmap for details. ## downloads the data library("rpx") px1 <- PXDataset("PXD000001") mzf <- pxget(px1, 7) ## reads the data ms <- openMSfile(mzf) hd <- header(ms) ## a set of spectra of interest: MS1 spectra eluted ## between 30 and 35 minutes retention time ms1 <- which(hd$msLevel == 1)
rtsel <- hd$retentionTime[ms1] / 60 > 30 & hd$retentionTime[ms1] / 60 < 35

## the map
M <- MSmap(ms, ms1[rtsel], 521, 523, .005, hd, zeroIsNA = TRUE)

mrda <- dir(system.file(package = "MSnbase", dir = "extdata"),
full.names = TRUE, pattern = "M.rda$") mrda2 <- dir(system.file(package = "MSnbase", dir = "extdata"), full.names = TRUE, pattern = "M2.rda$")

M


The M map object can be rendered as a heatmap with plot, as shown on figure \@ref(fig:mapheat).

plot(M, aspect = 1, allTicks = FALSE)


One can also render the data in 3 dimension with the plot3D function, as show on figure \@ref(fig:map3d).

plot3D(M)


To produce figure \@ref(fig:map3d2), we create a second MSmap object containing the first two MS1 spectra of the first map (object M above) and all intermediate MS2 spectra and display $m/z$ values between 100 and 1000.

i <- ms1[which(rtsel)][1]
j <- ms1[which(rtsel)][2]
M2 <- MSmap(ms, i:j, 100, 1000, 1, hd)

M2

plot3D(M2)


## MS Spectra {#sec:specplots}

Spectra can be plotted individually or as part of (subset) experiments with the plot method. Full spectra can be plotted (using full=TRUE), specific reporter ions of interest (by specifying with reporters with reporters=iTRAQ4 for instance) or both (see figure \@ref(fig:spectrumPlot)).

plot(sp, reporters = iTRAQ4, full = TRUE)

## bsasel <- fData(itraqdata)$ProteinAccession == "BSA" bsasel <- 1:3  It is also possible to plot all spectra of an experiment (figure \@ref(fig:msnexpPlot)). Lets start by subsetting the itraqdata experiment using the protein accession numbers included in the feature metadata, and keep the r sum(bsasel) from the BSA protein. sel <- fData(itraqdata)$ProteinAccession == "BSA"
bsa <- itraqdata[sel]
bsa
as.character(fData(bsa)$ProteinAccession)  These can then be visualised together by plotting the MSnExp object, as illustrated on figure \@ref(fig:msnexpPlot). plot(bsa, reporters = iTRAQ4, full = FALSE) + theme_gray(8)  Customising your plots The r Biocpkg("MSnbase") plot methods have a logical plot parameter (default is TRUE), that specifies if the plot should be printed to the current device. A plot object is also (invisibly) returned, so that it can be saved as a variable for later use or for customisation. r Biocpkg("MSnbase") uses the \CRANpkg{ggplot2} package to generate plots, which can subsequently easily be customised. More details about \CRANpkg{ggplot2} can be found in [@ggplot2] (especially chapter 8) and on http://had.co.nz/ggplot2/. Finally, if a plot object has been saved in a variable p, it is possible to obtain a summary of the object with summary(p). To view the data frame used to generate the plot, use p$data.

## MS Chromatogram {#sec:chromplots}

Chromatographic data can be plotted using the plot method which, in contrast to the plot method for Spectrum classes, uses R base graphics. The plot method is implemented for Chromatogram and MChromatograms classes. The latter plots all chromatograms for the same $m/z$-rt range of all files in an experiment (i.e. for one row in the MChromatograms object) into one plot.

plot(mtof_bpc)


# Tandem MS identification data {#sec:id}

Typically, identification data is produced by a search engine and serialised to disk in the mzIdentML (or mzid) file format. This format can be parsed by openIDfile from the r Biocpkg("mzR") package or mzID from the r Biocpkg("mzID") package. The MSnbase package relies on the former (which is faster) and offers a simplified interface by converting the dedicated identification data objects into data.frames.

library("msdata")
f <- "TMT_Erwinia_1uLSike_Top10HCD_isol2_45stepped_60min_01-20141210.mzid"
idf <- msdata::ident(full.names = TRUE, pattern = f)
str(iddf)


The spectra along the rows are duplicated when the PSM can be assigned to multiple proteins, such as

iddf[grep("scan=5291", iddf[, "spectrumID"]),
c("spectrumID", "sequence", "DatabaseAccess")]


of when there are multiple modifications in a PSM, such as

iddf[grep("scan=4936", iddf[, "spectrumID"]),
c("spectrumID", "sequence", "modName", "modLocation")]


At this stage, it is useful to perform some exploratory data analysis and visualisation on the identification data. For example

table(iddf$isDecoy) table(iddf$chargeState)

library("ggplot2")
ggplot(data = iddf, aes(x = MS.GF.RawScore, colour = isDecoy)) +
geom_density() +
facet_wrap(~chargeState)


The filterIdentificationDataFrame function can be used to remove - PSMs that match decoy entries - PSMs of rank > 1 - PSMs that match non-proteotypic proteins

iddf <- filterIdentificationDataFrame(iddf)


This data.frame can be now be further reduced so that individual rows represent unique spectra, which can be done with the reduce method.

iddf2 <- reduce(iddf, key = "spectrumID")


This reduces the number of rows from r nrow(iddf) to r nrow(iddf2).

The first duplicated spectrum mentioned above is now unique as is matched a decoy protein that was filtered out with filterIdentificationDataFrame.

iddf2[grep("scan=5291", iddf2[, "spectrumID"]),
c("spectrumID", "sequence", "DatabaseAccess")]


The matches to multiple modification in the same peptide are now combined into a single row and documented as semicolon-separated values.

iddf2[grep("scan=4936", iddf2[, "spectrumID"]),
c("spectrumID", "sequence", "modName", "modLocation")]


This is the form that is used when combined to raw data, as described in the next section.

r Biocpkg("MSnbase") is able to integrate identification data from mzIdentML [@Jones2012] files.

We first load two example files shipped with the r Biocpkg("MSnbase") containing raw data (as above) and the corresponding identification results respectively. The raw data is read with the readMSData, as demonstrated above. As can be seen, the default feature data only contain spectra numbers. More data about the spectra is of course available in an MSnExp object, as illustrated in the previous sections. See also ?pSet and ?MSnExp for more details.

## find path to a mzXML file
quantFile <- dir(system.file(package = "MSnbase", dir = "extdata"),
full.name = TRUE, pattern = "mzXML$") ## find path to a mzIdentML file identFile <- dir(system.file(package = "MSnbase", dir = "extdata"), full.name = TRUE, pattern = "dummyiTRAQ.mzid") ## create basic MSnExp msexp <- readMSData(quantFile, verbose = FALSE) head(fData(msexp), n = 2)  The addIdentificationData method takes an MSnExp instance (or an MSnSet instance storing quantitation data, see section \@ref(sec:quant)) as first argument and one or multiple mzIdentML file names (as a character vector) as second one^[The identification data can also be passed as dedicated identification objects such as mzRident from the r Biocpkg("mzR") package or mzID from thr r Biocpkg("mzID") package, or as a data.frame - see ?addIdentifionData for details.] and updates the MSnExp feature data using the identification data read from the mzIdentML file(s). msexp <- addIdentificationData(msexp, id = identFile) head(fData(msexp), n = 2)  Finally we can use idSummary to summarise the percentage of identified features per quantitation/identification pairs. idSummary(msexp)  When identification data is present, and hence peptide sequences, one can annotation fragment peaks on the MS2 figure by passing the peptide sequence to the plot method. itraqdata2 <- pickPeaks(itraqdata, verbose=FALSE) i <- 14 s <- as.character(fData(itraqdata2)[i, "PeptideSequence"])  plot(itraqdata2[[i]], s, main = s)  The fragment ions are calculated with the calculateFragments, described in section \@ref(sec:calcfrag). ## Filtering identification data One can remove the features that have not been identified using removeNoId. This function uses by default the pepseq feature variable to search the presence of missing data (NA values) and then filter these non-identified spectra. fData(msexp)$sequence
msexp <- removeNoId(msexp)
fData(msexp)$sequence idSummary(msexp)  Similarly, the removeMultipleAssignment method can be used to filter out non-unique features, i.e. that have been assigned to protein groups with more than one member. This function uses by default the nprot feature variable. Note that removeNoId and removeMultipleAssignment methods can also be called on MSnExp instances. ## Calculate Fragments {#sec:calcfrag} r Biocpkg("MSnbase") is able to calculate theoretical peptide fragments via calculateFragments. calculateFragments("ACEK", type = c("a", "b", "c", "x", "y", "z"))  It is also possible to match these fragments against an Spectrum2 object. pepseq <- fData(msexp)$sequence[1]
calculateFragments(pepseq, msexp[[1]], type=c("b", "y"))


# Quality control {#sec:qc}

The current section is not executed dynamically for package size and processing time constrains. The figures and tables have been generated with the respective methods and included statically in the vignette for illustration purposes.

r Biocpkg("MSnbase") allows easy and flexible access to the data, which allows to visualise data features to assess it's quality. Some methods are readily available, although many QC approaches will be experiment specific and users are encourage to explore their data.

The plot2d method takes one MSnExp instance as first argument to produce retention time vs. precursor MZ scatter plots. Points represent individual MS2 spectra and can be coloured based on precursor charge (with second argument z="charge"), total ion count (z="ionCount"), number of peaks in the MS2 spectra z="peaks.count") or, when multiple data files were loaded, file z="file"), as illustrated on the next figure. The lower right panel is produced for only a subset of proteins. See the method documentation for more details.

{#fig:plot2d}

The plotDensity method illustrates the distribution of several parameters of interest (see figure below). Similarly to plot2d, the first argument is an MSnExp instance. The second is one of precursor.mz, peaks.count or ionCount, whose density will be plotted. An optional third argument specifies whether the x axes should be logged.

{#fig:plotDensity}

The plotMzDelta method^[The code to generate the histograms has been contributed by Guangchuang Yu.] implements the $m/z$ delta plot from [@Foster11] The $m/z$ delta plot illustrates the suitability of MS2 spectra for identification by plotting the $m/z$ differences of the most intense peaks. The resulting histogram should optimally shown outstanding bars at amino acid residu masses. More details and parameters are described in the method documentation (?plotMzDelta). The next figure has been generated using the PRIDE experiment 12011, as in [@Foster11].

{#fig:plotMzDelta}

In section \@ref(sec:incompdissoc), we illustrate how to assess incomplete reporter ion dissociation.

# Raw data processing {#sec:rawprocessing}

## Cleaning spectra {#sec:clean}

There are several methods implemented to perform basic raw data processing and manipulation. Low intensity peaks can be set to 0 with the removePeaks method from spectra or whole experiments. The intensity threshold below which peaks are removed is defined by the t parameter. t can be specified directly as a numeric. The default value is the character "min", that will remove all peaks equal to the lowest non null intensity in any spectrum. We observe the effect of the removePeaks method by comparing total ion count (i.e. the total intensity in a spectrum) with the ionCount method before (object itraqdata) and after (object experiment) for spectrum X55. The respective spectra are shown on figure \@ref(fig:spectrum-clean-plot).

experiment <- removePeaks(itraqdata, t = 400, verbose = FALSE)
ionCount(itraqdata[["X55"]])
ionCount(experiment[["X55"]])

p1 <- plot(itraqdata[["X55"]], full = TRUE, plot = FALSE) + theme_gray(5)
p2 <- plot(experiment[["X55"]], full = TRUE, plot = FALSE) + theme_gray(5)
grid.newpage()
pushViewport(viewport(layout = grid.layout(1, 2)))
vplayout <- function(x, y)
viewport(layout.pos.row = x, layout.pos.col = y)
print(p1,vp=vplayout(1,1))
print(p2,vp=vplayout(1,2))


Unlike the name might suggest, the removePeaks method does not actually remove peaks from the spectrum; they are set to 0. This can be checked using the peaksCount method, that returns the number of peaks (including 0 intensity peaks) in a spectrum. To effectively remove 0 intensity peaks from spectra, and reduce the size of the data set, one can use the clean method. The effect of the removePeaks and clean methods are illustrated on figure \@ref(fig:preprocPlot).

peaksCount(itraqdata[["X55"]])
peaksCount(experiment[["X55"]])
experiment <- clean(experiment, verbose = FALSE)
peaksCount(experiment[["X55"]])

int <- c(0,1,1,3,1,1,0,0,0,1,3,7,3,1,0)
mz <- c(113.9,114.0,114.05,114.1,114.15,114.2,114.25,
114.3,114.35,114.4,114.42,114.48,114.5,114.55,114.6)
ppsp <- new("Spectrum2",intensity=int,mz=mz,centroided=FALSE)
p1 <- plot(ppsp, full = TRUE, plot = FALSE) + theme_gray(5) +
geom_point(size=3,alpha=I(1/3)) +
geom_hline(yintercept=3,linetype=2) +
ggtitle("Original spectrum")
p2 <- plot(removePeaks(ppsp,t=3), full=TRUE, plot = FALSE) +
theme_gray(5) +
geom_point(size=3,alpha=I(1/3)) +
geom_hline(yintercept=3,linetype=2) +
ggtitle("Peaks < 3 removed")
p3 <- plot(clean(removePeaks(ppsp,t=3)), full = TRUE, plot = FALSE) +
theme_gray(5) +
geom_point(size=3,alpha=I(1/3)) +
geom_hline(yintercept=3,linetype=2) +
ggtitle("Peaks < 3 removed and cleaned")

grid.newpage()
pushViewport(viewport(layout = grid.layout(3, 1)))
print(p1, vp=vplayout(1, 1))
print(p2, vp=vplayout(2, 1))
print(p3, vp=vplayout(3, 1))


## Focusing on specific MZ values {#sec:trim}

Another useful manipulation method is trimMz, that takes as parameters and MSnExp (or a Spectrum) and a numeric mzlim. MZ values smaller then min(mzlim) or greater then max(mzmax) are discarded. This method is particularly useful when one wants to concentrate on a specific MZ range, as for reporter ions quantification, and generally results in substantial reduction of data size. Compare the size of the full trimmed experiment to the original r sz Mb.

range(mz(itraqdata[["X55"]]))
experiment <- filterMz(experiment, mzlim = c(112,120))
range(mz(experiment[["X55"]]))
experiment


As can be seen above, all processing performed on the experiment is recorded and displayed as integral part of the experiment object.

## Spectrum processing {#sec:specproc}

MSnExp and Spectrum2 instances also support standard MS data processing such as smoothing and peak picking, as described in the smooth and pickPeak manual pages. The methods that either single spectra of experiments, process the spectrum/spectra, and return a updated, processed, object. The implementations originate from the \CRANpkg{MALDIquant} package [@Gibb2012].

# MS2 isobaric tagging quantitation {#sec:isoquant}

## Reporter ions quantitation {#sec:quant}

Quantitation is performed on fixed peaks in the spectra, that are specified with an ReporterIons object. A specific peak is defined by it's expected mz value and is searched for within mz $\pm$ width. If no data is found, NA is returned.

mz(iTRAQ4)
width(iTRAQ4)

int <- c(0, 1, 1, 3, 1, 1, 0)
mz <- c(113.9, 114.0, 114.05, 114.1, 114.15, 114.2, 114.25)
ssp <- new("Spectrum2", intensity = int, mz = mz, centroided = FALSE)
p <- plot(ssp, full = TRUE, plot = FALSE)
p <- p + theme_gray(5)


The quantify method takes the following parameters: an MSnExp experiment, a character describing the quantification method, the reporters to be quantified and a strict logical defining whether data points ranging outside of mz $\pm$ width should be considered for quantitation. Additionally, a progress bar can be displaying when setting the verbose parameter to TRUE. Three quantification methods are implemented, as illustrated on figure \@ref(fig:quantitationPlot). Quantitation using sum sums all the data points in the peaks to produce, for this example, r quantify(ssp, iTRAQ4[1], method = "sum")[[1]], whereas method max only uses the peak's maximum intensity, r quantify(ssp, iTRAQ4[1], method = "max")[[1]]. Trapezoidation calculates the area under the peak taking the full with into account (using strict = FALSE gives r round(quantify(ssp, iTRAQ4[1], method = "trap", strict = FALSE)[[1]], 3)) or only the width as defined by the reporter (using strict = TRUE gives r round(quantify(ssp, iTRAQ4[1], method = "trap", strict = TRUE)[[1]],3)). See ?quantify for more details.

grid.newpage()
pushViewport(viewport(layout = grid.layout(2, 2)))
vplayout <- function(x, y)
viewport(layout.pos.row = x, layout.pos.col = y)
print(p + ggtitle("Quantitation using 'sum'") +
geom_point(size = 3, alpha = I(1/3), colour = "red"),
vp = vplayout(1, 1))
print(p + ggtitle("Quantitation using 'max'") +
geom_point(aes(x = 114.1, y = 3),
alpha = I(1/18),
colour = "red", size = 3),
vp = vplayout(1, 2))
print(p +
ggtitle("Trapezoidation and strict=FALSE") +
geom_polygon(alpha = I(1/5), fill = "red"),
vp = vplayout(2, 1))
print(p +
ggtitle("Trapezoidation and strict=TRUE") +
geom_polygon(aes(x = c(NA, 114.05, 114.05, 114.1, 114.15, 114.15, NA),
y = c(NA, 0, 1, 3, 1, 0, NA)), fill = "red", alpha = I(1/5)),
vp = vplayout(2,2))


The quantify method returns MSnSet objects, that extend the well-known eSet class defined in the r Biocpkg("Biobase") package. MSnSet instances are very similar to ExpressionSet objects, except for the experiment meta-data that captures MIAPE specific information. The assay data is a matrix of dimensions $n \times m$, where $m$ is the number of features/spectra originally in the MSnExp used as parameter in quantify and $m$ is the number of reporter ions, that can be accessed with the exprs method. The meta data is directly inherited from the MSnExp instance.

qnt <- quantify(experiment,
method = "trap",
reporters = iTRAQ4,
strict = FALSE,
verbose = FALSE)
qnt


The next figure illustrates the quantitation of the TMT 10-plex isobaric tags using the quantify method and the TMT10 reporter instance. The data on the $x$ axis has been quantified using method = "max" and centroided data (as generated using ProteoWizard's msconvert with vendor libraries' peak picking); on the $y$ axis, the quantitation method was trapezoidation and strict = TRUE (that's important for TMT 10-plex) and the profile data. We observe a very good correlation.

{#fig:tmt10}

If no peak is detected for a reporter ion peak, the respective quantitation value is set to NA. In our case, there is r sum(is.na(exprs(qnt))) such case in row r which(is.na(exprs(qnt))) %% nrow(qnt). We will remove the offending line using the filterNA method. The pNA argument defines the percentage of accepted missing values per feature. As we do not expect any missing peaks, we set it to be 0 (which is also the detault value).

table(is.na(qnt))
qnt <- filterNA(qnt, pNA = 0)
sum(is.na(qnt))


The filtering criteria for filterNA can also be defined as a pattern of columns that can have missing values and columns that must not exhibit any. See ?filterNA for details and examples.

The infrastructure around the MSnSet class allows flexible filtering using the [ sub-setting operator. Below, we mimic the behaviour of filterNA(, pNA = 0) by calculating the row indices that should be removed, i.e. those that have at least one NA value and explicitly remove these rows. This method allows one to devise and easily apply any filtering strategy.

whichRow <- which(is.na((qnt))) %% nrow(qnt)
qnt <- qnt[-whichRow, ]


See also the plotNA method to obtain a graphical overview of the completeness of a data set.

## Importing quantitation data {#sec:io2}

If quantitation data is already available as a spreadsheet, it can be imported, along with additional optional feature and sample (pheno) meta data, with the readMSnSet function. This function takes the respective text-based spreadsheet (comma- or tab-separated) file names as argument to create a valid MSnSet instance.

Note that the quantitation data of MSnSet objects can also be exported to a text-based spreadsheet file using the write.exps method.

r Biocpkg("MSnbase") also supports the mzTab format\footnote{https://github.com/HUPO-PSI/mzTab}, a light-weight, tab-delimited file format for proteomics data. mzTab files can be read into R with readMzTabData to create and MSnSet instance.

See the MSnbase-io vignette for a general overview of r Biocpkg("MSnbase")'s input/ouput capabilites.

## Importing chromatographic data from SRM/MRM experiments {#sec:io3}

Data from SRM/MRM experiments can be imported from mzML files using the readSRMData function. The mzML files are expected to contain chromatographic data for the same precursor and product m/z values. The function returns a MChromatograms object that arranges the data in a two-dimensional array, each column representing the data of one file (sample) and each row the chromatographic data for the same polarity, precursor and product m/z. In the example code below we load a single SRM file using readSRMData.

fl <- proteomics(full.names = TRUE, pattern = "MRM")

srm


The precursor and product m/z values can be extracted with the precursorMz and productMz functions. These functions always return a matrix, each row providing the lower and upper m/z value of the isolation window (in most cases minimal and maximal m/z will be identical).

head(precursorMz(srm))



Single peak adjustment In certain cases, peak intensities need to be adjusted as a result of peak interferance. For example, the $+1$ peak of the phenylalanine (F, Phe) immonium ion (with $m/z$ 120.03) inteferes with the 121.1 TMT reporter ion. Below, we calculate the relative intensity of the +1 peaks compared to the main peak using the r Biocpkg("Rdisop") package.

library(Rdisop)
## Phenylalanine immonium ion
Fim <- getMolecule("C8H10N")
getMass(Fim)
isotopes <- getIsotope(Fim)
F1 <- isotopes[2, 2]
F1


If desired, one can thus specifically quantify the F immonium ion in the MS2 spectrum, estimate the intensity of the +1 ion (r round(F1,4)% of the F peak) and substract this calculated value from the 121.1 TMT reporter intensity.

The above principle can also be generalised for a set of overlapping peaks, as described below.

Reporter ions purity correction Impurities in the reporter reagents can also bias the results and can be corrected when manufacturers provide correction coefficients. These generally come as percentages of each reporter ion that have masses differing by -2, -1, +1 and +2 Da from the nominal reporter ion mass due to isotopic variants. The purityCorrect method applies such correction to MSnSet instances. It also requires a square matrix as second argument, impurities, that defines the relative percentage of reporter in the quantified each peak. See ?purityCorrect for more details.

impurities <- matrix(c(0.929, 0.059, 0.002, 0.000,
0.020, 0.923, 0.056, 0.001,
0.000, 0.030, 0.924, 0.045,
0.000, 0.001, 0.040, 0.923),
nrow = 4)
qnt.crct <- purityCorrect(qnt, impurities)


The makeImpuritiesMatrix can be used to create impurity matrices. It opens a rudimentary spreadsheet that can be directly edited.

# Processing quantitative data {#sec:qproc}

## Data imputation {#sec:imp}

A set of imputation methods are available in the impute method: it takes an MSnSet instance as input, the name of the imputation method to be applied (one of r paste(imputeMethods(), collapse=", ")), possible additional parameters and returns an updated for MSnSet without any missing values. Below, we apply a deterministic minimum value imputation on the naset example data:

## an example MSnSet containing missing values
data(naset)
table(is.na(naset))
## number of NAs per protein
table(fData(naset)$nNA) x <- impute(naset, "min") processingData(x) table(is.na(x))  As described in more details in [@Lazar:2016], there are two types of mechanisms resulting in missing values in LC/MSMS experiments. • Missing values resulting from absence of detection of a feature, despite ions being present at detectable concentrations. For example in the case of ion suppression or as a result from the stochastic, data-dependent nature of the MS acquisition method. These missing value are expected to be randomly distributed in the data and are defined as missing at random (MAR) or missing completely at random (MCAR). • Biologically relevant missing values, resulting from the absence of the low abundance of ions (below the limit of detection of the instrument). These missing values are not expected to be randomly distributed in the data and are defined as missing not at random (MNAR). MAR and MCAR values can be reasonably well tackled by many imputation methods. MNAR data, however, requires some knowledge about the underlying mechanism that generates the missing data, to be able to attempt data imputation. MNAR features should ideally be imputed with a left-censor (for example using a deterministic or probabilistic minimum value) method. Conversely, it is recommended to use hot deck methods (for example nearest neighbour, maximum likelihood, etc) when data are missing at random. x <- impute(naset, "zero") exprs(x)[exprs(x) != 0] <- 1 suppressPackageStartupMessages(library("gplots")) heatmap.2(exprs(x), col = c("lightgray", "black"), scale = "none", dendrogram = "none", trace = "none", keysize = 0.5, key = FALSE, RowSideColors = ifelse(fData(x)$randna, "orange", "brown"),
ColSideColors = rep(c("steelblue", "darkolivegreen"), each = 8))


It is anticipated that the identification of both classes of missing values will depend on various factors, such as feature intensities and experimental design. Below, we use perform mixed imputation, applying nearest neighbour imputation on the r sum(fData(naset)$randna) features that are assumed to contain randomly distributed missing values (if any) (yellow on figure \@ref(fig:miximp)) and a deterministic minimum value imputation on the r sum(!fData(naset)$randna) proteins that display a non-random pattern of missing values (brown on figure \@ref(fig:miximp)).

x <- impute(naset, method = "mixed",
randna = fData(naset)$randna, mar = "knn", mnar = "min") x  Please read ?impute for a description of the different methods. ## Normalisation {#sec:norm} A MSnSet object is meant to be compatible with further downstream packages for data normalisation and statistical analysis. There is also a normalise (also available as normalize) method for expression sets. The method takes and instance of class MSnSet as first argument, and a character to describe the method to be used: • quantiles: Applies quantile normalisation [@Bolstad03] as implemented in the normalize.quantiles function of the r Biocpkg("preprocessCore") package. • quantiles.robust: Applies robust quantile normalisation [@Bolstad03] as implemented in the normalize.quantiles.robust function of the r Biocpkg("preprocessCore") package. • vsn: Applies variance stabilisation normalization [@Huber2002] as implemented in the vsn2 function of the r Biocpkg("vsn") package. • max: Each feature's reporter intensity is divided by the maximum of the reporter ions intensities. • sum: Each feature's reporter intensity is divided by the sum of the reporter ions intensities. See ?normalise for more methods. A scale method for MSnSet instances, that relies on the base::scale function. qnt.max <- normalise(qnt, "max") qnt.sum <- normalise(qnt, "sum") qnt.quant <- normalise(qnt, "quantiles") qnt.qrob <- normalise(qnt, "quantiles.robust") qnt.vsn <- normalise(qnt, "vsn")  The effect of these are illustrated on figure \@ref(fig:normPlot) and figure \@ref(fig:cvPlot) reproduces figure 3 of [@Karp2010] that described the application of vsn on iTRAQ reporter data. .plot <- function(x, ttl = NULL) { boxplot(exprs(x), main = ifelse(is.null(ttl), processingData(x)@processing[2], ttl), cex.main = .8, cex.lab = .5, cex.axis = .5, cex = .8) grid() } oldmar <- par()$mar
par(mfrow=c(3,2),mar=c(2.9,2.9,2.9,1))
.plot(qnt, ttl = "Non-normalised data")
.plot(qnt.max, ttl = "Maximum")
.plot(qnt.sum, ttl = "Sum")
.plot(qnt.quant, ttl = "Quantile")
.plot(qnt.qrob, ttl = "Robust quantile")
.plot(qnt.vsn, ttl = "vsn")

sd1 <- apply(log2(exprs(qnt))+10,1,sd)
mn1 <- apply(log2(exprs(qnt))+10,1,mean)
cv1 <- sd1/mn1
sd2 <- apply(exprs(qnt.vsn)+10,1,sd)
mn2 <- apply(exprs(qnt.vsn)+10,1,mean)
cv2 <- sd2/mn2
dfr <- rbind(data.frame(rank=order(mn1),cv=cv1,norm="raw"),
data.frame(rank=order(mn2),cv=cv2,norm="vsn"))
library("zoo")
## rmed1 <- rollapply(cv1,7,function(x) median(x,na.rm=TRUE))
## rmed2 <- rollapply(cv2,7,function(x) median(x,na.rm=TRUE))
##
## Calling directly rollapply.zoo to make it zoo_1.6-4 compatible.
## The above requires zoo >= 1.7-0, which is as of 15 March 2011
## not yet available on CRAN (only on r-forge).
rmed1 <- zoo:::rollapply.zoo(cv1,7,function(x) median(x,na.rm=TRUE))
rmed2 <- zoo:::rollapply.zoo(cv2,7,function(x) median(x,na.rm=TRUE))
dfr2 <-
rbind(data.frame(x=seq(1,70,by=(70/length(rmed1))),
y=rmed1,norm="raw"),
data.frame(x=seq(1,70,by=(70/length(rmed1))),
y=rmed2,norm="vsn"))

p <- qplot(rank,cv,data=dfr,col=norm) +
geom_line(data=dfr2,aes(x=x,y=y,colour=norm)) +
theme_gray(7)

print(p)


Note that it is also possible to normalise individual spectra or whole MSnExp experiments with the normalise method using the max method. This will rescale all peaks between 0 and 1. To visualise the relative reporter peaks, one should this first trim the spectra using method trimMz as illustrated in section \@ref(sec:rawprocessing), then normalise the MSnExp with normalise using method="max" as illustrated above and plot the data using plot (figure \@ref(fig:msnexpNormPlot)).

p <- plot(normalise(experiment[bsasel], "max"),
reporters = iTRAQ4, full = FALSE, plot = FALSE)
p <- p + theme_gray(7)

print(p)


Additional dedicated normalisation method are available for MS2 label-free quantitation, as described in section \@ref(sec:lf) and in the quantify documentation.

# Feature aggregation {#sec:feataggregation}

The above quantitation and normalisation has been performed on quantitative data obtained from individual spectra. However, the biological unit of interest is not the spectrum but the peptide or the protein. As such, it is important to be able to summarise features that belong to a same group, i.e. spectra from one peptide, peptides that originate from one protein, or directly combine all spectra that have been uniquely associated to one protein.

r Biocpkg("MSnbase") provides one function, combineFeatures, that allows to aggregate features stored in an MSnSet using build-in or user defined summary function and return a new MSnSet instance. The three main arguments are described below. Additional details can be found in the method documentation.

gb <- fData(qnt)$ProteinAccession  combineFeatures's first argument, object, is an instance of class MSnSet, as has been created in the section \@ref(sec:quant) for instance. The second argument, groupBy, is a factor than has as many elements as there are features in the MSnSet object argument. The features corresponding to the groupBy levels will be aggregated so that the resulting MSnSet output will have length(levels(groupBy)) features. Here, we will combine individual MS2 spectra based on the protein they originate from. As shown below, this will result in r length(table(gb)) new aggregated features. gb <- fData(qnt)$ProteinAccession
table(gb)
length(unique(gb))


The third argument, method, defined how to combine the features. Predefined functions are readily available and can be specified as strings (method="mean", method="median", method="sum", method="weighted.mean" or method="medianpolish" to compute respectively the mean, media, sum, weighted mean or median polish of the features to be aggregated). Alternatively, is is possible to supply user defined functions with method=function(x) { ... }. We will use the median here.

qnt2 <- combineFeatures(qnt, groupBy = gb, method = "median")
qnt2


Of interest is also the iPQF spectra-to-protein summarisation method, which integrates peptide spectra characteristics and quantitative values for protein quantitation estimation. See ?iPQF and references therein for details.

# Label-free MS2 quantitation {#sec:lf}

## Peptide counting

Note that if samples are not multiplexed, label-free MS2 quantitation by spectral counting is possible using r Biocpkg("MSnbase"). Once individual spectra have been assigned to peptides and proteins (see section \@ref(sec:id)), it becomes straightforward to estimate protein quantities using the simple peptide counting method, as illustrated in section \@ref(sec:feataggregation).

sc <- quantify(msexp, method = "count")
## lets modify out data for demonstration purposes
fData(sc)$DatabaseAccess[1] <- fData(sc)$DatabaseAccess[2]
fData(sc)$DatabaseAccess sc <- combineFeatures(sc, groupBy = fData(sc)$DatabaseAccess,
method = "sum")
exprs(sc)


Such count data could then be further analyses using dedicated count methods (originally developed for high-throughput sequencing) and directly available for MSnSet instances in the r Biocpkg("msmsTests") Bioconductor package.

## Spectral counting and intensity methods

The spectral abundance factor (SAF) and the normalised form (NSAF) [@Paoletti2006] as well as the spectral index (SI) and other normalised variations (SI$_{GI}$ and SI$_N$) [@Griffin2010] are also available. Below, we illustrate how to apply the normalised SI$_N$ to the experiment containing identification data produced in section \@ref(sec:id).

The spectra that did not match any peptide have already been remove with the removeNoId method. As can be seen in the following code chunk, the first spectrum could not be matched to any single protein. Non-identified spectra and those matching multiple proteins are removed automatically prior to any label-free quantitation. Once can also remove peptide that do not match uniquely to proteins (as defined by the nprot feature variable column) with the removeMultipleAssignment method.

fData(msexp)[, c("DatabaseAccess", "nprot")]


Note that the label-free methods implicitely apply feature aggregation (section \@ref(sec:feataggregation)) and normalise (section \@ref(sec:norm)) the quantitation values based on the total sample intensity and or the protein lengths (see [@Paoletti2006] and [@Griffin2010] for details).

Let's now proceed with the quantitation using the quantify, as in section \@ref(sec:quant), this time however specifying the method of interest, SIn (the reporters argument can of course be ignored here). The required peptide-protein mapping and protein lengths are extracted automatically from the feature meta-data using the default accession and length feature variables.

siquant <- quantify(msexp, method = "SIn")
processingData(siquant)
exprs(siquant)


Other label-free methods can be applied by specifiying the appropriate method argument. See ?quantify for more details.

# Spectra comparison

## Plotting two spectra

r Biocpkg("MSnbase") provides functionality to compare spectra against each other. The first notable function is plot. If two Spectrum2 objects are provided plot will draw two plots: the upper and lower panel contain respectively the first and second spectrum. Common peaks are drawn in a slightly darker colour.

centroided <- pickPeaks(itraqdata, verbose = FALSE)
(k <- which(fData(centroided)[, "PeptideSequence"] == "TAGIQIVADDLTVTNPK"))
mzk <- precursorMz(centroided)[k]
zk <- precursorCharge(centroided)[k]
mzk * zk
plot(centroided[[k[1]]], centroided[[k[2]]])


## Comparison metrics

Currently r Biocpkg("MSnbase") supports three different metrics to compare spectra against each other: common to calculate the number of common peaks, cor to calculate the Pearson correlation and dotproduct to calculate the dot product. See ?compareSpectra to apply other arbitrary metrics.

compareSpectra(centroided[[2]], centroided[[3]],
fun = "common")
compareSpectra(centroided[[2]], centroided[[3]],
fun = "cor")
compareSpectra(centroided[[2]], centroided[[3]],
fun = "dotproduct")


compareSpectra supports MSnExp objects as well.

compmat <- compareSpectra(centroided, fun="cor")
compmat[1:10, 1:5]


Below, we illustrate how to compare a set of spectra using a hierarchical clustering.

plot(hclust(as.dist(compmat)))


# Quantitative assessment of incomplete dissociation {#sec:incompdissoc}

Quantitation using isobaric reporter tags assumes complete dissociation between the reporter group (red on the figure below), balance group (blue) and peptide (the peptide reactive group is drawn in green). However, incomplete dissociation does occur and results in an isobaric tag (i.e reporter and balance groups) specific peaks.

{#fig:itraqchem}

r Biocpkg("MSnbase") provides, among others, a ReporterIons object for iTRAQ 4-plex that includes the 145 peaks, called iTRAQ5. This can then be used to quantify the experiment as show in section \@ref(sec:quant) to estimate incomplete dissociation for each spectrum.

iTRAQ5
incompdiss <- quantify(itraqdata,
method = "trap",
reporters = iTRAQ5,
strict = FALSE,
verbose = FALSE)


Figure \@ref(fig:incompdissPlot) compares these intensities for the whole experiment.

dfr <- cbind(expand.grid("feature" = featureNames(incompdiss),
"reporters" = sub("iTRAQ5\\.", "",
sampleNames(incompdiss)),
stringsAsFactors=FALSE),
"intensity" = as.vector(exprs(incompdiss)))
repsum <- rowSums(exprs(incompdiss)[, 1:4])
dfr2 <- data.frame(iTRAQ1to4 = repsum, iTRAQ5 = exprs(incompdiss)[,5])
p1 <- ggplot(data = dfr, aes(x = reporters,y = log10(intensity))) +
geom_boxplot() +
theme_gray(6)
p2 <- ggplot(data = dfr2, aes(x = log10(iTRAQ1to4), y = log10(iTRAQ5))) +
geom_point(alpha = I(1/2)) +
geom_abline(intercept = 0, slope = 1, linetype = "dotted") +
stat_smooth(method = lm, se = FALSE) +
xlab(label = expression(log[10]~sum~114~to~117)) +
ylab(label = expression(log[10]~145)) +
theme_gray(6)
grid.newpage()
pushViewport(viewport(layout = grid.layout(1, 2)))
print(p1, vp = vplayout(1, 1))
print(p2, vp = vplayout(1, 2))


# Combining MSnSet instances {#sec:combine}

Combining mass spectrometry runs can be done in two different ways depending on the nature of these runs. If the runs represent repeated measures of identical samples, for instance multiple fractions, the data has to be combine along the row of the quantitation matrix: all the features (along the rows) represent measurements of the same set of samples (along the columns). In this situation, described in section \@ref(sec:comb1), two experiments of dimensions $n_1$ (rows) by $m$ (columns and $n_2$ by $m$ will produce a new experiment of dimensions $n_1 + n_2$ by $m$.

When however, different sets of samples have been analysed in different mass spectrometry runs, the data has to be combined along the columns of the quantitation matrix: some features will be shared across experiments and should thus be aligned on a same row in the new data set, whereas unique features to one experiment should be set as missing in the other one. In this situation, described in section \@ref(sec:comb2), two experiments of dimensions $n_1$ by $m_1$ and $n_2$ by $m_2$ will produce a new experiment of dimensions $unique_{n_1} + unique_{n_2} + shared_{n_1, n_2}$ by $m_1 + m_2$. The two first terms of the first dimension will be complemented by NA values.

Default MSnSet feature names (X1, X2, ...) and sample names (iTRAQ4.114, iTRAQ4.115, iTRAQ4.116, ...) are not informative. The features and samples of these anonymous quantitative data-sets should be updated before being combined, to guide how to meaningfully merge them.

## Combining identical samples {#sec:comb1}

To simulate this situation, let us use quantiation data from the itraqdata object that is provided with the package as experiment 1 and the data from the rawdata MSnExp instance created at the very beginning of this document. Both experiments share the same default iTRAQ 4-plex reporter names as default sample names, and will thus automatically be combined along rows.

exp1 <- quantify(itraqdata, reporters = iTRAQ4,
verbose = FALSE)
sampleNames(exp1)
centroided(rawdata) <- FALSE
exp2 <- quantify(rawdata, reporters = iTRAQ4,
verbose = FALSE)
sampleNames(exp2)


It important to note that the features of these independent experiments share the same default anonymous names: X1, X2, X3, ..., that however represent quantitation of distinct physical analytes. If the experiments were to be combined as is, it would result in an error because data points for the same feature name (say X1) and the same sample name (say iTRAQ4.114) have different values. We thus first update the feature names to explicitate that they originate from different experiment and represent quantitation from different spectra using the convenience function updateFeatureNames. Note that updating the names of one experiment would suffice here.

head(featureNames(exp1))
exp1 <- updateFeatureNames(exp1)
exp2 <- updateFeatureNames(exp2)


The two experiments now share the same sample names and have different feature names and will be combined along the row. Note that all meta-data is correctly combined along the quantitation values.

exp12 <- combine(exp1, exp2)
dim(exp1)
dim(exp2)
dim(exp12)


## Combine different samples {#sec:comb2}

Lets now create two MSnSets from the same raw data to simulate two different independent experiments that share some features. As done previously (see section \@ref(sec:feataggregation)), we combine the spectra based on the proteins they have been identified to belong to. Features can thus naturally be named using protein accession numbers. Alternatively, if peptide sequences would have been used as grouping factor in combineFeatures, then these would be good feature name candidates.

set.seed(1)
i <- sample(length(itraqdata), 35)
j <- sample(length(itraqdata), 35)
exp1 <- quantify(itraqdata[i], reporters = iTRAQ4,
verbose = FALSE)
exp2 <- quantify(itraqdata[j], reporters = iTRAQ4,
verbose = FALSE)
exp1 <- droplevels(exp1)
exp2 <- droplevels(exp2)
table(featureNames(exp1) %in% featureNames(exp2))
exp1 <- combineFeatures(exp1,
groupBy = fData(exp1)$ProteinAccession) exp2 <- combineFeatures(exp2, groupBy = fData(exp2)$ProteinAccession)


The droplevels drops the unused featureData levels. This is required to avoid passing absent levels as groupBy in combineFeatures. Alternatively, one could also use factor(fData(exp1)\$ProteinAccession) as groupBy argument. The feature names are updated automatically by combineFeatures, using the groupBy argument. Proper feature names, reflecting the nature of the features (spectra, peptides or proteins) is critical when multiple experiments are to be combined, as this is done using common features as defined by their names (see below). Sample names should also be updated to replace anonymous reporter names with relevant identifiers; the individual reporter data is stored in the phenoData and is not lost. A convenience function updateSampleNames is provided to append the MSnSet's variable name to the already defined names, although in general, biologically relevant identifiers are preferred. sampleNames(exp1) exp1 <- updateSampleNames(exp1) sampleNames(exp1) sampleNames(exp1) <- c("Ctrl1", "Cond1", "Ctrl2", "Cond2") sampleNames(exp2) <- c("Ctrl3", "Cond3", "Ctrl4", "Cond4")  At this stage, it is not yet possible to combine the two experiments, because their feature data is not compatible yet; they share the same feature variable labels, i.e. the feature data column names (r paste(head(fvarLabels(exp1), n = 3), collapse=", "), ...), but the part of the content is different because the original data was (in particular all the spectrum centric data: identical peptides in different runs will have different retention times, precursor intensities, ...). Feature data with identical labels (columns in the data frame) and names (row in the data frame) are expected to have the same data and produce an error if not conform. stopifnot(all(fvarLabels(exp1) == fvarLabels(exp2))) fData(exp1)["BSA", 1:4] fData(exp2)["BSA", 1:4]  Instead of removing these identical feature data columns, one can use a second convenience function, updateFvarLabels, to update feature labels based on the experiements variable name and maintain all the metadata. exp1 <- updateFvarLabels(exp1) exp2 <- updateFvarLabels(exp2) head(fvarLabels(exp1)) head(fvarLabels(exp2))  It is now possible to combine exp1 and exp2, including all the meta-data, with the combine method. The new experiment will contain the union of the feature names of the individual experiments with missing values inserted appropriately. exp12 <- combine(exp1, exp2) dim(exp12) pData(exp12) exprs(exp12)[25:28, ] exp12  In summary, when experiments with different samples need to be combined (along the columns), one needs to (1) clarify the sample names using updateSampleNames or better manually, for biological relevance and (2) update the feature data variable labels with updateFvarLabels. The individual experiments (there can be more than 2) can then easily be combined with the combine method while retaining the meta-data. If runs for the same sample (different fractions for example) need to be combines, one needs to (1) differentiate the feature provenance with updateFeatureNames prior to use combine. ## Splitting and unsplitting MSnSet instances {#sec:split} A single MSnSet can also be split along the features/rows or samples/columns using the split method and a factor defining the splitting groups, resulting in an instance of class MSnSetList: data(dunkley2006) head(pData(dunkley2006)) split(dunkley2006, dunkley2006$replicate)
## or, defining the appropriate annotation variable name
dun <- split(dunkley2006, "replicate")


Above, we split along the columns/samples, but the function would equally work with a factor of length equal to the number of rows of the MSnSet (or a feature variable name) to split along the rows/features.

Finally, the effect of split can be reverted by unsplit.

dun2 <- unsplit(dun, pData(dunkley2006)$replicate) compareMSnSets(dunkley2006, dun2)  See ?MSnSetList for more details about the class, split and unsplit and comments about storing multiple assays pertaining the same experiment. ## Averaging MSnSet instances {#sec:avg} It is sometimes useful to average a set of replicated experiments to facilitate their visualisation. This can be easily achieved with the averageMSnSet function, which takes a list of valid MSnSet instances as input and creates a new object whose expression values are an average of the original values. A value of dispersion (disp) and a count of missing values (nNA) is recorded in the feature metadata slot. The average and dispersion are computed by default as the median and (non-parametric) coefficient of variation (see ?npcv for details), although this can easily be parametrised, as described in ?averageMSnSet. The next code chunk illustrates the averaging function using three replicated experiments from [@Tan2009] available in the r Biocpkg("pRolocdata") package. library("pRolocdata") data(tan2009r1) data(tan2009r2) data(tan2009r3) msnl <- MSnSetList(list(tan2009r1, tan2009r2, tan2009r3)) avgtan <- averageMSnSet(msnl) head(exprs(avgtan)) head(fData(avgtan)$disp)
head(fData(avgtan)$nNA)  We are going to visualise the average data on a principle component (PCA) plot using the plot2D function from the r Biocpkg("pRoloc") package [@Gatto2014]. In addition, we are going to use the measure of dispersion to highlight averages with high variability by taking, for each protein, the maximum observed dispersion in the 4 samples. Note that in the default implementation, dispersions estimated from a single measurement (i.e. that had 2 missing values in our example) are set to 0; we will set these to the overal maximum observed dispersion. disp <- rowMax(fData(avgtan)$disp)
disp[disp == 0] <- max(disp)
range(disp)
library("pRoloc")
plot2D(avgtan, cex = 3 * disp)


# MS^E^ data processing {#sec:mse}

r Biocpkg("MSnbase") can also be used for MS^E^ data independent acquisition from Waters instrument. The MS^E^ pipeline depends on the Bioconductor r Biocpkg("synapter") package [@Bond2013] that produces MSnSet instances for indvidual acquisitions. The r Biocpkg("MSnbase") infrastructure can subsequently be used to further combine experiments, as shown in section \@ref(sec:comb2) and apply top3 quantitation using the topN method.

# Session information {-}

sessionInfo()


# References {-}

## Try the MSnbase package in your browser

Any scripts or data that you put into this service are public.

MSnbase documentation built on Jan. 23, 2021, 2 a.m.