%\VignetteEngine{knitr} %\VignetteIndexEntry{Base Functions and Classes for MS-based Proteomics} %\VignetteKeywords{Mass Spectrometry, MS, MSMS, Proteomics, Infrastructure, Bioinformatics, quantitative } %\VignettePackage{MSnbase-demo} \documentclass[12pt,a4paper,english]{scrartcl} \usepackage{amsmath,amsfonts,amssymb} \usepackage{tikz} \usepackage{hyperref} \usepackage[authoryear,round]{natbib} \usepackage[auth-sc]{authblk} \usepackage{setspace} \onehalfspacing % caption formatting \setcapindent{0em} \setkomafont{captionlabel}{\sffamily\bfseries} \setkomafont{caption}{\sffamily} \newcommand{\R}{\texttt{R} } \newcommand{\code}[1]{{\texttt{#1}}} \newcommand{\Rfunction}[1]{{\texttt{#1}}} \newcommand{\Robject}[1]{{\texttt{#1}}} \newcommand{\Rpackage}[1]{{\mbox{\normalfont\textsf{#1}}}} \newcommand{\email}[1]{\href{mailto:#1}{\normalfont\texttt{#1}}} %% colors \definecolor{Red}{rgb}{0.7,0,0} \definecolor{Blue}{rgb}{0,0,0.8} \usepackage{geometry} \geometry{verbose, tmargin = 2.5cm, bmargin = 2.5cm, lmargin = 3.5cm, rmargin = 3.5cm} \usepackage{hyperref} \usepackage{breakurl} \hypersetup{% pdfusetitle, bookmarks = {true}, bookmarksnumbered = {true}, bookmarksopen = {true}, bookmarksopenlevel = 2, unicode = {true}, breaklinks = {false}, hyperindex = {true}, colorlinks = {true}, linktocpage = {true}, plainpages = {false}, linkcolor = {Blue}, citecolor = {Blue}, urlcolor = {Red}, pdfstartview = {Fit}, pdfpagemode = {UseOutlines}, pdfview = {XYZ null null null} } \author{ Laurent Gatto\thanks{\email{lg390@cam.ac.uk}} } \author{ Sebastian Gibb } \affil{ Computational Proteomics Unit\\ University of Cambridge, UK } \begin{document} \title{ \Rpackage{MSnbase}: labelled and label-free MS2 data pre-processing, visualisation and quantification. } \maketitle %% Abstract and keywords %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \vskip 0.3in minus 0.1in \hrule \begin{abstract} This vignette describes the functionality implemented in the \Rpackage{MSnbase} package. \Rpackage{MSnbase} aims at (1) facilitating the import, processing, visualisation and quantification of mass spectrometry data into the \R environment \citep{Rstat} by providing specific data classes and methods and (2) enabling the utilisation of throughput-high data analysis pipelines provided by the Bioconductor \citep{Gentleman2004} project. \end{abstract} \textit{Keywords}: Mass Spectrometry (MS), proteomics, infrastructure, quantitative. \vskip 0.1in minus 0.05in \hrule \vskip 0.2in minus 0.1in %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \newpage \begin{small} \tableofcontents \end{small} \newpage <>= suppressPackageStartupMessages(library("ggplot2")) suppressPackageStartupMessages(library("MSnbase")) suppressPackageStartupMessages(library("zoo")) suppressPackageStartupMessages(require("Rdisop")) suppressPackageStartupMessages(require("pRolocdata")) suppressPackageStartupMessages(require("pRoloc")) library("grid") library("reshape2") @ <>= library(knitr) opts_chunk$set(fig.align = 'center', fig.show = 'hold', par = TRUE, prompt = TRUE, comment = NA) options(replace.assign = TRUE, width = 68) ## see ?evaluate::evaluate for details opts_knit$set(error=TRUE) ## knit_hooks$set(par = function(before, options, envir) { ## if (before && options$fig.show != 'none') ## par(mar = c(4,4,.1,.1), ## cex.lab = .95, ## cex.axis = .9, ## mgp = c(2,.7,0), ## tcl = -.3) ## }) @ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \input{Foreword.tex} \input{Bugs.tex} \newpage %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Introduction}\label{sec:intro} \Rpackage{MSnbase} \citep{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. \Rpackage{MSnbase} also aims at being compatible with the infrastructure implemented in Bioconductor, in particular \Rpackage{Biobase}. As such, classes developed specifically for proteomics mass spectrometry data are based on the \Robject{eSet} and \Robject{Expression} classes. The main goal is to assure seamless compatibility with existing meta data structure, accessor methods and normalisation techniques. This vignette illustrates \Rpackage{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 \texttt{MSnbase-development} vignette. \input{NoteAboutSpeedAndMemory.tex} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Data structure and content}\label{sec:data} \subsection{Importing experiments}\label{sec:io} \Rpackage{MSnbase} is able to import raw MS data stored in one of the \texttt{XML}-based formats as well as peak lists in the \texttt{mfg} format\footnote{%% Mascot Generic Format -- \url{http://www.matrixscience.com/help/data\_file\_help.html\#GEN}} \paragraph{Raw data} The \texttt{XML}-based formats, \texttt{mzXML} \citep{Pedrioli2004}, \texttt{mzData} \citep{Orchard2007} and \texttt{mzML} \citep{Martens2010} can be imported with the \Rfunction{readMSData} function, as illstrated below (see \Rfunction{?readMSData} for more details). <>= file <- dir(system.file(package = "MSnbase", dir = "extdata"), full.names = TRUE, pattern = "mzXML$") rawdata <- readMSData(file, msLevel = 2, verbose = FALSE) @ %% $ Either MS1 or MS2 spectra can be loaded at a time by setting the \texttt{msLevel} parameter accordingly. In this document, we will use the \Robject{itraqdata} data set, provided with \Rpackage{MSnbase}. It includes feature metadata, accessible with the \Rfunction{fData} accessor. The metadata includes identification data for the \Sexpr{length(itraqdata)} MS2 spectra. \paragraph{Peak lists} 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 \texttt{mgf} format can be imported with the function \Rfunction{readMgfData} (see \Rfunction{?readMgfData} for details) to create experiment objects. Experiments or individual spectra can be exported to an \texttt{mgf} file with the \Rfunction{writeMgfData} methods (see \Rfunction{?writeMgfData} for details and examples). \bigskip \paragraph{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} (page \pageref{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 using \Rfunction{readMSData} on a 32-cores servers with 128 Gb of RAM. It took about 90 minutes to read the 14 uncentroided \texttt{mzXML} raw files (4.9 Gb on disk in total) and create a 3.3 Gb raw data object (an \Robject{MSnExp} instance, see next section). Quantitation of 9 reporter ions (\Robject{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 and analysed on a standard laptop computer. Since verions 1.13.5, parallel support is provided by the \Rpackage{BiocParallel} and various backends including multicore (forking), simple networf network of workstations (SNOW) using sockets, forking or MPI among others. \bigskip See also section \ref{sec:io2} to import quantitative data stored in spreadsheets into \R for further processing using \Rpackage{MSnbase}. The \texttt{MSnbase-io} vignette gives a general overview of \Rpackage{MSnbase}'s input/ouput capabilites. \subsection{MS experiments}\label{sec:msnexp} Raw data is contained in \Robject{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: \begin{itemize} \item Information about the raw data, i.e. the spectra. \item Specific information about the experiment processing\footnote{%% this part will be automatically updated when the object is modified with it's \textit{ad hoc} methods, as illustrated later} and package version. This slot can be accessed with the \Rfunction{processingData} method. \item 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 \Robject{eSet} class and are described in more details in their respective manual pages. See \texttt{?MSnExp} and references therein for additional background information. The experiment meta data associated with an \Robject{MSnExp} experiment is of class \Robject{MIAPE}. It stores general information about the experiment as well as MIAPE (Minimum Information About a Proteomics Experiment) information \citep{Taylor2007, Taylor2008}. This meta-data can be accessed with the \Rfunction{experimentData} method. When available, a summary of MIAPE-MS data can be printed with the \Rfunction{msInfo} method. See \texttt{?MIAPE} for more details. \end{itemize} \subsection{Spectra objects}\label{sec:spectra} The raw data is composed of the \Sexpr{length(itraqdata)} MS spectra. The spectra are named individually (\Sexpr{paste(paste(head(featureNames(itraqdata)),collapse=", "),", ...",sep="")}) and stored in a \Robject{environment}. They can be accessed individually with \texttt{itraqdata[["X1"]]} or \texttt{itraqdata[[1]]}, or as a list with \texttt{spectra(itraqdata)}. As we have loaded our experiment specifying \texttt{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: \Rfunction{precursorCharge} for the precursor charge, \Rfunction{rtime} for the retention time, \Rfunction{mz} for the MZ values, \Rfunction{intensity} for the intensities, ... see the \Robject{Spectrum}, \Robject{Spectrum1} and \Robject{Spectrum2} manuals for more details. <>= peaksCount(sp) head(peaksCount(itraqdata)) rtime(sp) head(rtime(itraqdata)) @ \subsection{Reporter ions}\label{sec:reporterions} Reporter ions are defined with the \Robject{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}). \Robject{ReporterIons} instances are required to quantify reporter peaks in \Robject{MSnExp} experiments. Instances for the most commonly used isobaric tags like iTRAQ 4-plex and 8-plex and TMT tags are already defined in \Rpackage{MSnbase}. See \texttt{?ReporterIons} for details about how to generate new \Robject{ReporterIons} objects. <>= iTRAQ4 @ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Plotting raw data}\label{sec:plotting} \subsection{Default plots}\label{sec:defaultplots} Spectra can be plotted individually or as part of (subset) experiments with the \Robject{plot} method. Full spectra can be plotted (using \texttt{full=TRUE}), specific reporter ions of interest (by specifying with reporters with \texttt{reporters=iTRAQ4} for instance) or both (see figure \ref{fig:spectrum-plot}). \begin{figure}[!ht] <>= plot(sp, reporters = iTRAQ4, full = TRUE) @ \caption{Raw MS2 spectrum with details about reporter ions.} \label{fig:spectrum-plot} \end{figure} <>= ## bsasel <- fData(itraqdata)$ProteinAccession == "BSA" bsasel <- 1:3 @ %% $ It is also possible to plot all spectra of an experiment (figure \ref{fig:msnexp-plot}). Lets start by subsetting the \Robject{itraqdata} experiment using the protein accession numbers included in the feature metadata, and keep the \Sexpr{sum(bsasel)} from the \textit{BSA} protein. <>= sel <- fData(itraqdata)$ProteinAccession == "BSA" bsa <- itraqdata[sel] bsa as.character(fData(bsa)$ProteinAccession) @ %% Lets start by extracting all spectra that have the same precursor MZ value than \Robject{sp} %% into a separate experiment, using the \Rfunction{extractPrecSpectra} method. %% This method takes an \Robject{MSnExp} experiment and a precursor MZ value as parameters. %% <>= %% exp2 <- extractPrecSpectra(raw.experiment,precursorMz(sp)) %% exp2 %% @ These can then be visualised together by plotting the \Robject{MSnExp} object, as illustrated on figure \ref{fig:msnexp-plot}. \begin{figure}[p] <>= plot(bsa, reporters = iTRAQ4, full = FALSE) + theme_gray(8) @ \caption{Experiment-wide raw MS2 spectra. The y-axes of the individual spectra are automatically rescaled to the same range. See section \ref{sec:norm} to rescale peaks identically. } \label{fig:msnexp-plot} \end{figure} \subsection{Customising your plots}\label{sec:customplots} The \Rpackage{MSnbase} \Rfunction{plot} methods have a logical \Robject{plot} parameter (default is \Robject{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. \Rpackage{MSnbase} uses the \Rpackage{ggplot2} package to generate plots, which can subsequently easily be customised. More details about \Rpackage{ggplot2} can be found in \cite{ggplot2} (especially chapter 8) and on \href{http://had.co.nz/ggplot2/}{http://had.co.nz/ggplot2/}. Finally, if a plot object has been saved in a variable \Robject{p}, it is possible to obtain a summary of the object with \Rfunction{summary(p)}. To view the data frame used to generate the plot, use \Rfunction{p@data}. \clearpage %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Tandem MS identification data}\label{sec:id} \subsection{Adding identification data} \Rpackage{MSnbase} is able to integrate identification data from \texttt{mzIdentML} \citep{Jones2012} files. We first load two example files shipped with the \Rpackage{MSnbase} containing raw data (as above) and the corresponding identification results respectively. The raw data is read with the \Rfunction{readMSData}, as demonstrated above. As can be seen, the default feature data only contain spectra numbers\footnote{More data about the spectra is of course available in an \Robject{MSnExp} object, as illustrated in the previous sections. See also \Rfunction{?pSet} and \Rfunction{?MSnExp} for more details.}. <>= ## find path to a mzXML file quantFile <- dir(system.file(package = "MSnbase", dir = "extdata"), full.name = TRUE, pattern = "mzXML$") ## find path to a mzIdentML file identFile <- dir(system.file(package = "MSnbase", dir = "extdata"), full.name = TRUE, pattern = "mzid$") ## create basic MSnExp msexp <- readMSData(quantFile, verbose = FALSE) head(fData(msexp), n = 2) @ The \Rfunction{addIdentificationData} method takes an \Robject{MSnExp} instance (or an \Robject{MSnSet} instance storing quantitation data, see section \ref{sec:quant}) as first argument and one or multiple \texttt{mzIdentML} file names (as a character vector) as second one and updates the \Robject{MSnExp} feature data using the identification data read from the \texttt{mzIdentML} file(s). <>= ## add identification information msexp <- addIdentificationData(msexp, filenames = identFile, verbose = FALSE) head(fData(msexp), n = 2) @ Finally we can use \Rfunction{idSummary} to summarise the percentage of identified features per quantitation/identification pairs. <>= idSummary(msexp) @ \subsection{Filtering identification data} One can remove the features that have not been identified using \Rfunction{removeNoId}. This function uses by default the \code{pepseq} feature variable to search the presence of missing data (\code{NA} values) and then filter these non-identified spectra. <>= fData(msexp)$pepseq msexp <- removeNoId(msexp) fData(msexp)$pepseq idSummary(msexp) @ Similarly, the \Rfunction{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 \code{nprot} feature variable. \bigskip Note that \Rfunction{removeNoId} and \Rfunction{removeMultipleAssignment} methods can also be called on \code{MSnExp} instances. \subsection{Calculate Fragments} \Rpackage{MSnbase} is able to calculate theoretical peptide fragments via \Rfunction{calculateFragments}. <>= calculateFragments("ACEK", type=c("a", "b", "c", "x", "y", "z")) @ It is also possible to match these fragments against an \Robject{Spectrum2} object. <>= pepseq <- fData(msexp)$pepseq[1] calculateFragments(pepseq, msexp[[1]], type=c("b", "y")) @ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Quality control}\label{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. \bigskip \Rpackage{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 \Rfunction{plot2d} method takes one \Robject{MSnExp} instance as first argument to produce retention time \textit{vs.} precursor MZ scatter plots. Points represent individual MS2 spectra and can be coloured based on precursor charge (with second argument \Robject{z="charge"}), total ion count (\Robject{z="ionCount"}), number of peaks in the MS2 spectra \Robject{z="peaks.count"}) or, when multiple data files were loaded, file \Robject{z="file"}), as illustrated on figure \ref{fig:plot2d}. The lower right panel is produced for only a subset of proteins. See the method documentation for more details. \begin{figure}[htb] \includegraphics[width=\linewidth]{./plot2d-figure.png} \caption{ Illustration of the \Rfunction{plot2d} output. } \label{fig:plot2d} \end{figure} The \Rfunction{plotDensity} method illustrates the distribution of several parameters of interest (see figure \ref{fig:plotDensity}). Similarly to \Rfunction{plot2d}, the first argument is an \Robject{MSnExp} instance. The second is one of \Robject{precursor.mz}, \Robject{peaks.count} or \Robject{ionCount}, whose density will be plotted. An optional third argument specifies whether the x axes should be logged. \begin{figure}[htb] \includegraphics[width=\linewidth]{./plotDensity-figure.png} \caption{ Illustration of the \Rfunction{plotDensity} output. } \label{fig:plotDensity} \end{figure} %% xtable(preprocSelectionTable(velos)) %% % latex table generated in R 2.14.0 by xtable 1.5-6 package %% % Thu May 19 10:22:29 2011 %% \begin{table}[ht] %% \begin{center} %% \begin{tabular}{rr} %% \hline %% & x \\ %% \hline %% 1 & 6836 \\ %% 2 & 59 \\ %% 3 & 5 \\ %% \hline %% \end{tabular} %% \end{center} %% \end{table} %% matplot on iTRAQ5 quantitation (see allquant code chunk) \bigskip The \Rfunction{plotMzDelta} method\footnote{The code to generate the histograms has been contributed by Guangchuang Yu from Jinan University, China.} implements the M/Z delta plot from \cite{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 (\Rfunction{?plotMzDelta}). Figure \ref{fig:plotMzDelta-pride12011} has been generated using the PRIDE experiment 12011, as in \cite{Foster11}. \begin{figure}[htb] \begin{center} \includegraphics[width=.7\linewidth]{./plotMzDelta-pride12011.pdf} \caption{ Illustration of the \Rfunction{plotMzDelta} output for the PRIDE experiment 12011, as in figure 4A from \cite{Foster11}. } \label{fig:plotMzDelta-pride12011} \end{center} \end{figure} \bigskip In section \ref{sec:incompdissoc} on page \pageref{sec:incompdissoc}, we illustrate how to assess incomplete reporter ion dissociation. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Data processing}\label{sec:processing} \subsection{Cleaning spectra}\label{sec:clean} There are several methods implemented to perform basic data manipulation. Low intensity peaks can be set to 0 with the \Rfunction{removePeaks} method from spectra or whole experiments. The intensity threshold below which peaks are removed is defined by the \Robject{t} parameter. \Robject{t} can be specified directly as a numeric. The default value is the character \texttt{"min"}, that will remove all peaks equal to the lowest non null intensity in any spectrum. We observe the effect of the \Rfunction{removePeaks} method by comparing total ion count (i.e. the total intensity in a spectrum) with the \Rfunction{ionCount} method before (object \Robject{itraqdata}) and after (object \Robject{experiment}) for spectrum \texttt{X55}. The respective spectra are shown on figure \ref{fig:spectrum-clean-plot} (page \pageref{fig:spectrum-clean-plot}). <>= experiment <- removePeaks(itraqdata, t = 400, verbose = FALSE) ## total ion current ionCount(itraqdata[["X55"]]) ionCount(experiment[["X55"]]) @ \begin{figure}[!ht] <>= 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)) @ \caption{Same spectrum before (left) and after setting peaks <= 400 to 0.} \label{fig:spectrum-clean-plot} \end{figure} Unlike the name might suggest, the \Rfunction{removePeaks} method does not actually remove peaks from the spectrum; they are set to 0. This can be checked using the \Rfunction{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 \Rfunction{clean} method. The effect of the \Rfunction{removePeaks} and \Rfunction{clean} methods are illustrated on figure \ref{fig:preproc} on page \pageref{fig:preproc}. <>= ## number of peaks 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") @ \begin{figure}[p] <>= 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)) @ \caption{This figure illustrated the effect or the \Rfunction{removePeaks} and \Rfunction{clean} methods. The left-most spectrum displays two peaks, of max height 3 and 7 respectively. The middle spectrum shows the result of calling \Rfunction{removePeaks} with argument \Robject{t=3}, which sets all data points of the first peak, whose maximum height is smaller or equal to \Robject{t} to 0. The second peak is unaffected. Calling \Rfunction{clean} after \Rfunction{removePeaks} effectively deletes successive 0 intensities from the spectrum, as shown on the right plot. } \label{fig:preproc} \end{figure} \subsection{Focusing on specific MZ values}\label{sec:trim} Another useful manipulation method is \Rfunction{trimMz}, that takes as parameters and \Robject{MSnExp} (or a \Robject{Spectrum}) and a numeric \Robject{mzlim}. MZ values smaller then \texttt{min(mzlim)} or greater then \texttt{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 \Sexpr{sz} Mb. \label{trimMz-example} <>= range(mz(itraqdata[["X55"]])) experiment <- trimMz(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. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{MS$^2$ isobaric tagging quantitation}\label{sec:isoquant} \subsection{Reporter ions quantitation}\label{sec:quant} Quantitation is performed on fixed peaks in the spectra, that are specified with an \Robject{ReporterIons} object. A specific peak is defined by it's expected \Robject{mz} value and is searched for within \Robject{mz} $\pm$ \Robject{width}. If no data is found, \Robject{NA} is returned. <>= mz(iTRAQ4) width(iTRAQ4) @ The \Rfunction{quantify} method takes the following parameters: an \Robject{MSnExp} experiment, a character describing the quantification \Robject{method}, the \Robject{reporters} to be quantified and a \Robject{strict} logical defining whether data points ranging outside of \Robject{mz} $\pm$ \Robject{width} should be considered for quantitation. Additionally, a progress bar can be displaying when setting the \Robject{verbose} parameter to \Robject{TRUE}. Three quantification methods are implemented, as illustrated on figure \ref{fig:quant-methods}: \Robject{trapezoidation} returns the area under the peak of interest, \Robject{max} returns the apex of the peak and \Robject{sum} returns the sum of all intensities of the peak. See \texttt{?quantify} for more details. <>= 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) @ \begin{figure}[p] <>= 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)) @ \caption{The different quantitation methods are illustrated above. Quantitation using \Robject{sum} sums all the data points in the peaks to produce, for this example, \Sexpr{quantify(ssp,iTRAQ4[1],method="sum")[[1]]}, whereas method \Robject{max} only uses the peak's maximum intensity, \Sexpr{quantify(ssp,iTRAQ4[1],method="max")[[1]]}. \Robject{Trapezoidation} calculates the area under the peak taking the full with into account (using \Robject{strict=FALSE} gives \Sexpr{round(quantify(ssp,iTRAQ4[1],method="trap",strict=FALSE)[[1]],3)}) or only the width as defined by the reporter (using \Robject{strict=TRUE} gives \Sexpr{round(quantify(ssp,iTRAQ4[1],method="trap",strict=TRUE)[[1]],3)}). } \label{fig:quant-methods} \end{figure} \bigskip The \Rfunction{quantify} method returns \Robject{MSnSet} objects, that extend the well-known \Robject{eSet} class defined in the \Rpackage{Biobase} package. \Robject{MSnSet} instances are very similar to \Robject{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 \Robject{MSnExp} used as parameter in \Robject{quantify} and $m$ is the number of reporter ions, that can be accessed with the \Rfunction{exprs} method. The meta data is directly inherited from the \Robject{MSnExp} instance. <>= qnt <- quantify(experiment, method = "trap", reporters = iTRAQ4, strict = FALSE, verbose = FALSE) qnt head(exprs(qnt)) @ If no peak is detected for a reporter ion peak, the respective quantitation value is set to \texttt{NA}. In our case, there is \Sexpr{sum(is.na(exprs(qnt)))} such case in row \Sexpr{which(is.na(exprs(qnt))) %% nrow(qnt)}. We will remove the offending line using the \Rfunction{filterNA} method. The \Robject{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 \Rfunction{filterNA} can also be defined as a pattern of columns that can have missing values and columns that must not exhibit any. See \Rfunction{?filterNA} for details and examples. The infrastructure around the \Robject{MSnSet} class allows flexible filtering using the \Rfunction{[} sub-setting operator. Below, we mimic the behaviour of \Rfunction{filterNA(, pNA = 0)} by calculating the row indices that should be removed, i.e. those that have at least on \Robject{NA} value and explicitly remove these row. 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 \Rfunction{plotNA} method to obtain a graphical overview of the completeness of a data set. \subsection{Importing quantitation data}\label{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 \Rfunction{readMSnSet} function. This function takes the respective text-based spreadsheet (comma- or tab-separated) file names as argument to create a valid \Robject{MSnSet} instance. Note that the quantitation data of \Robject{MSnSet} objects can also be exported to a text-based spreadsheet file using the \Rfunction{write.exps} method. \Rpackage{MSnbase} also supports the \texttt{mzTab} format\footnote{\url{http://code.google.com/p/mztab/}}, a light-weight, tab-delimited file format for proteomics data. \texttt{mzTab} files can be read into \R with \Rfunction{readMzTabData} to create and \Robject{MSnSet} instance. \Robject{MSnSet} objects can also be exported to \texttt{mzTab} with the \Rfunction{writeMzTabData} function. \bigskip See the \texttt{MSnbase-io} vignette for a general overview of \Rpackage{MSnbase}'s input/ouput capabilites. \subsection{Peak adjustments}\label{sec:purcor} \paragraph{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 \Rpackage{Rdispo} 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 (\Sexpr{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. \paragraph{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 \Rfunction{purityCorrect} method applies such correction to \Robject{MSnSet} instances. It also requires a square matrix as second argument, \Robject{impurities}, that defines the relative percentage of reporter in the quantified each peak. See \texttt{?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) head(exprs(qnt)) head(exprs(qnt.crct)) @ The \Rfunction{makeImpuritiesMatrix} can be used to create impurity matrices. It opens a rudimentary spreadsheet that can be directly edited. \subsection{Normalisation}\label{sec:norm} A \Robject{MSnSet} object is meant to be compatible with further downstream packages for data normalisation and statistical analysis. There is also a \Rfunction{normalise} (also available as \Rfunction{normalize}) method for expression sets. The method takes and instance of class \Robject{MSnSet} as first argument, and a character to describe the \Robject{method} to be used: \begin{description} \item{\Robject{quantiles}} Applies quantile normalisation \citep{Bolstad03} as implemented in the \Rfunction{normalize.quantiles} function of the \Rpackage{preprocessCore} package. \item{\Robject{quantiles.robust}} Applies robust quantile normalisation \citep{Bolstad03} as implemented in the \Rfunction{normalize.quantiles.robust} function of the \Rpackage{preprocessCore} package. \item{\Robject{vsn}} Applies variance stabilisation normalization \citep{Huber2002} as implemented in the \Rfunction{vsn2} function of the \Rpackage{vsn} package. \item{\Robject{max}} Each feature's reporter intensity is divided by the maximum of the reporter ions intensities. \item{\Robject{sum}} Each feature's reporter intensity is divided by the sum of the reporter ions intensities. \end{description} See \Rfunction{?normalise} for more methods. A \Rfunction{scale} method for \Robject{MSnSet} instances, that relies on the \Rfunction{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:norm-plot} and figure \ref{fig:cv-plot} reproduces figure 3 of \cite{Karp2010} that described the application of vsn on iTRAQ reporter data. \begin{figure}[p] <>= .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") @ \caption{ Comparison of the normalisation \Robject{MSnSet} methods. Note that vsn also glog-transforms the intensities. } \label{fig:norm-plot} \end{figure} <>= 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) @ \begin{figure}[!ht] <>= print(p) @ \caption{CV versus signal intensity comparison for log2 and vsn transformed data. Lines indicate running CV medians.} \label{fig:cv-plot} \end{figure} Note that it is also possible to normalise individual spectra or whole \Robject{MSnExp} experiments with the \Rfunction{normalise} method using the \Robject{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 \Rfunction{trimMz} as illustrated in section \ref{sec:processing}, then normalise the \Robject{MSnExp} with \Rfunction{normalise} using \Robject{method="max"} as illustrated above and plot the data using \Rfunction{plot} (figure \ref{fig:msnexp-norm-plot}). <>= p <- plot(normalise(experiment[bsasel], "max"), reporters = iTRAQ4, full = FALSE, plot = FALSE) p <- p + theme_gray(7) @ \begin{figure}[p] <>= print(p) @ \caption{Experiment-wide normalised MS2 spectra. The y-axes of the individual spectra is now rescaled between 0 and 1 (highest peak), as opposed to figure \ref{fig:msnexp-plot}. } \label{fig:msnexp-norm-plot} \end{figure} \bigskip Additional dedicated normalisation method are available for MS$^2$ label-free quantitation, as described in section \ref{sec:lf} and in the \Rfunction{quantify} documentation. \section{Feature aggregation}\label{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. \Rpackage{MSnbase} provides one function, \Rfunction{combineFeatures}, that allows to aggregate features stored in an \Robject{MSnSet} using build-in or user defined summary function and return a new \Robject{MSnSet} instance. The three main arguments are described below. Additional details can be found in the method documentation. <>= gb <- fData(qnt)$ProteinAccession @ \Rfunction{combineFeatures}'s first argument, \Robject{object}, is an instance of class \Robject{MSnSet}, as has been created in the section \ref{sec:quant} for instance. The second argument, \Robject{groupBy}, is a \Robject{factor} than has as many elements as there are features in the \Robject{MSnSet} \Robject{object} argument. The features corresponding to the \Robject{groupBy} levels will be aggregated so that the resulting \Robject{MSnSet} output will have \Rfunction{length(levels(groupBy))} features. Here, we will combine individual MS2 spectra based on the protein they originate from. As shown below, this will result in \Sexpr{length(table(gb))} new aggregated features. <>= gb <- fData(qnt)$ProteinAccession table(gb) length(unique(gb)) @ The third argument, \Robject{fun}, defined how to combine the features. Predefined functions are readily available and can be specified as strings (\Robject{fun="mean"}, \Robject{fun="median"}, \Robject{fun="sum"}, \Robject{fun="weighted.mean"} or \Robject{fun="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 \Robject{fun=function(x) \{ ... \}}. We will use the \Robject{median} here. <>= qnt2 <- combineFeatures(qnt, groupBy = gb, fun = "median") qnt2 @ %% Other intended ways to define aggregation groups is to import %% identification data into R and collate it to the \Robject{MSnSet} %% feature data, and subsequently use peptide sequences or protein assignments %% to combine features. \section{Label-free MS$^2$ quantitation}\label{sec:lf} \subsection{Peptide counting} Note that if samples are not multiplexed, label-free MS$^2$ quantitation by spectral counting is possible using \Rpackage{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)$accession[1] <- fData(sc)$accession[2] fData(sc)$accession sc <- combineFeatures(sc, groupBy = fData(sc)$accession, fun = "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 \Robject{MSnSet} instances in the \Rpackage{msmsTests} Bioconductor package. \subsection{Spectral counting and intensity methods} The spectral abundance factor (SAF) and the normalised form (NSAF) \citep{Paoletti2006} as well as the spectral index (SI) and other normalised variations (SI$_{GI}$ and SI$_N$) \citep{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 \Rfunction{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 \code{nprot} feature variable column) with the \Rfunction{removeMultipleAssignment} method. <>= fData(msexp)[, c("accession", "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 \cite{Paoletti2006} and \cite{Griffin2010} for details). Let's now proceed with the quantitation using the \Rfunction{quantify}, as in section \ref{sec:quant}, this time however specifying the method of interest, \code{SIn} (the \code{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 \code{accession} and \code{length} feature variables. <>= siquant <- quantify(msexp, method = "SIn") processingData(siquant) exprs(siquant) @ Other label-free methods can be applied by specifiying the appropriate \Rfunction{method} argument. See \Rfunction{?quantify} for more details. \section{Spectra comparison} \subsection{Plotting two spectra} \Rpackage{MSnbase} provides functionality to compare spectra against each other. The first notable function is \Rfunction{plot}. If two \Robject{Spectrum2} objects are provided \Rfunction{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(msexp, verbose=FALSE) plot(centroided[[2]], centroided[[3]]) @ \subsection{Comparison metrics} Currently \Rpackage{MSnbase} supports three different metrics to compare spectra against each other: \code{common} to calculate the number of common peaks, \code{cor} to calculate the Pearson correlation and \code{dotproduct} to calculate the dot product. See \code{?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") @ \Rfunction{compareSpectra} supports \Robject{MSnExp} objects as well. <>= compareSpectra(centroided, fun="cor") @ Below, we illustrate how to compare a set of spectra using a hierarchical clustering. <>= dm <- as.dist(compareSpectra(centroided, fun="cor")) plot(hclust(dm)) @ \section{Quantitative assessment of incomplete dissociation} \label{sec:incompdissoc} Quantitation using isobaric reporter tags assumes complete dissociation between the reporter group (red on figure \ref{fig:itraqchem}), 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. \begin{figure}[h] \begin{center} \includegraphics[width=4cm]{./itraqchem.pdf} \caption{ iTRAQ 4-plex isobaric tags reagent consist of three parts: (1) a charged reporter group (MZ of 114, 115, 116 and 117) that is unique to each of the four reagents (red), (2) an uncharged mass balance group (28-31 Da) (blue)and (3) a peptide reactive group (NHS ester) that binds to the peptide. In case of incomplete dissociation, the reporter and balance groups produce a specific peaks at MZ 145. } \label{fig:itraqchem} \end{center} \end{figure} \Rpackage{MSnbase} provides, among others, a \Robject{ReporterIons} object for iTRAQ 4-plex that includes the 145 peaks, called \Robject{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) head(exprs(incompdiss)) @ Figure \ref{fig:incompdiss} compares these intensities for the whole experiment. \begin{figure}[!hbt] <>= dfr <- melt(exprs(incompdiss)) colnames(dfr) <- c("feature", "reporters", "intensity") dfr$reporters <- sub("iTRAQ5.", "", dfr$reporters) repsum <- apply(exprs(incompdiss), 1, function(x) sum(x[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)) @ \caption{Boxplot and scatterplot comparing intensities of the 4 reporter ions (or their sum, on the right) and the incomplete dissociation specific peak. } \label{fig:incompdiss} \end{figure} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Combining \Robject{MSnSet} instances}\label{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 \Robject{NA} values. Default \Robject{MSnSet} feature names (\Robject{X1}, \Robject{X2}, \ldots) and sample names (\Robject{iTRAQ4.114}, \Robject{iTRAQ4.115}, \Robject{iTRAQ4.116}, \ldots) 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. \subsection{Combining identical samples}\label{sec:comb1} To simulate this situation, let us use quantiation data from the \Robject{itraqdata} object that is provided with the package as experiment 1 and the data from the \Robject{rawdata} \Robject{MSnExp} instance created at the very beginning of this document. Both experiments share the \emph{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) 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, \ldots, 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 \emph{feature} name (say \Robject{X1}) and the same \emph{sample name} (say \Robject{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 \Rfunction{updateFeatureNames}. Note that updating the names of one experiment would suffice here. <>= head(featureNames(exp1)) exp1 <- updateFeatureNames(exp1) head(featureNames(exp1)) head(featureNames(exp2)) exp2 <- updateFeatureNames(exp2) head(featureNames(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) @ \subsection{Combine different samples}\label{sec:comb2} Lets now create two \Robject{MSnSet}s 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 \Rfunction{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) head(featureNames(exp1)) head(featureNames(exp2)) @ The \Rfunction{droplevels} drops the unused \Robject{featureData} levels. This is required to avoid passing absent levels as \Robject{groupBy} in \Rfunction{combineFeatures}. Alternatively, one could also use \Robject{factor(fData(exp1)\$ProteinAccession)} as \Robject{groupBy} argument. The feature names are updated automatically by \Rfunction{combineFeatures}, using the \Robject{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). \bigskip Sample names should also be updated to replace anonymous reporter names with relevant identifiers; the individual reporter data is stored in the \Robject{phenoData} and is not lost. A convenience function \Rfunction{updateSampleNames} is provided to append the \Robject{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 (\Sexpr{paste(head(fvarLabels(exp1), n = 3), collapse=", ")}, \ldots), 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, \ldots). 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, \Rfunction{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 \Robject{exp1} and \Robject{exp2}, including all the meta-data, with the \Rfunction{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 @ \bigskip In summary, when experiments with different samples need to be combined (along the columns), one needs to (1) clarify the sample names using \Rfunction{updateSampleNames} or better manually, for biological relevance and (2) update the feature data variable labels with \Rfunction{updateFvarLabels}. The individual experiments (there can be more than 2) can then easily be combined with the \Rfunction{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 \Rfunction{updateFeatureNames} prior to use \Rfunction{combine}. \subsection{Averaging \Robject{MSnSet} instances}\label{sec:avg} It is sometimes useful to average a set of replicated experiments to facilitate their visualisation. This can be easily achieved with the \Rfunction{averageMSnSet} function, which takes a list of valid \Robject{MSnSet} instances as input and creates a new object whose expression values are an average of the original values. A value of dispersion (\code{disp}) and a count of missing values (\code{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 \code{?npcv} for details), although this can easily be parametrised, as described in \code{?averageMSnSet}. The next code chunk illustrates the averaging function using three replicated experiments from \cite{Tan2009} available in the \Rpackage{pRolocdata} package. <>= library("pRolocdata") data(tan2009r1) data(tan2009r2) data(tan2009r3) avgtan <- averageMSnSet(list(tan2009r1, tan2009r2, tan2009r3)) 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 \Rfunction{plot2D} function from the \Rpackage{pRoloc} package \cite{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)$disp) disp[disp == 0] <- max(disp) range(disp) library("pRoloc") plot2D(avgtan, cex = 7 * disp) @ \section{MS$^E$ data processing}\label{sec:mse} \Rpackage{MSnbase} can also be used for MS$^E$ data independent acquisition from Waters instrument. The MS$^E$ pipeline depends on the Bioconductor \Rpackage{synapter} package \citep{Bond2013} that produces \Robject{MSnSet} instances for indvidual acquisitions. The \Rpackage{MSnbase} infrastructure can subsequently be used to further combine experiments, as shown in section \ref{sec:comb2} and apply \textit{top3} quantitation using the \Rfunction{topN} method. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% \section{Conclusion}\ref{sec:ccl} \section{Session information}\label{sec:sessionInfo} <>= toLatex(sessionInfo()) @ \bibliographystyle{plainnat} \bibliography{MSnbase} \end{document}