MSstats is an open-source R package for statistical relative quantification of proteins and peptides in global, targeted and data-independent proteomics. More information on MSstats
The MSstats Galaxy tool (version 4.0.0) allows the detection of differentially abundant proteins for label-free MS experiments with complex designs on data derived from open-source proteomics software available in Galaxy (e.g. MaxQuant, OpenMS, OpenSWATH). Processing functionalities such as log transformation, normalization, feature selection, missing value imputation and quantification are available as well.
Input data
Data in tabular or csv format, either in the 10-column MSstats format or the outputs of spectral processing tools such as MaxQuant, OpenSWATH
MSstats format: tabular file with 10 column either manually curated or other sources such as Swath2stats tool which is implemented in Pyprophet export in Galaxy. For manual curation: Names of headers are fixed but not case sensitive:
ProteinName: protein ID or peptide ID for peptide-level modeling and analysis; statistical analysis will be done separately for each unique label in this column
PeptideSequence: Amino acid sequence for each peptide. If the peptide sequences should be distinguished based on post-translational modifications, this column can be renamed to PeptideModifiedSequence.
PrecursorCharge: charge state of precursor.
FragmentIon: e.g. b4, y3, if unknown use a single value for all entries.
ProductCharge: charge state of product. If unknown use 0 for all entries.
IsotopeLabelType: This column indicates whether this measurement is based on the endogenous peptides (use “L”) or labeled reference peptides (use “H”).
Condition: For group comparison experiments, this column indicates groups of interest (such as “Disease” or “Control”). The name of the condition is not allowed to start with a number or contain any special characters. For time-course experiments, this column indicates time points (such as “T1”, “T2”, etc). If the experimental design contains both distinct groups of subjects and multiple time points per subject, this column should indicate a combination of these values (such as “Disease_T1”, “Disease_T2”, “Control_T1”, “Control_T2”, etc.).
BioReplicate: This column should contain a unique identifier for each biological replicate in the experiment. For example, in a clinical proteomic investigation this should be a unique patient id. Patients from distinct groups should have distinct ids. MSstats does not require the presence of technical replicates in the experiment. If the technical replicates are present, all samples or runs from a same biological replicate should have a same id. MSstats automatically detects the presence of technical replicates and accounts for them in the model-based analysis.
Run: This column contains the identifier of a mass spectrometry run. Each mass spectrometry run should have a unique identifier, regardless of the origin of the biological sample. In SRM experiments, if all the transitions of a biological or a technical replicate are split into multiple “methods” due to the technical limitations, each method should have a separate identifier. When processed by Skyline, distinct values of runs correspond to distinct input file names. It is possible to use the actual input file names as values in the column Run.
Intensity: This column should contain the quantified signal of a feature in a run without any transformation (in particular, no logarithm transform). The signals can be quantified as the peak height or the peak of area under curve. Any other quantitative representation of abundance can also be used.
Example file header:
proteinname peptidesequence precursorcharge fragmention productcharge P02768 DLGEENFK 3 y7 0 P02768 DLGEENFK 3 y8 0 P02768 ETYGEMADCCAK 2 b3 0 P02768 ETYGEMADCCAK 2 b4 0 ... ... ... ... ... isotopelabeltype condition bioreplicate run intensity L disease ReplA 1 4298.12 H disease ReplA 1 1974.59 L disease ReplA 1 7183.22 H disease ReplA 1 8467.58 ... ... ... ... ...MaxQuant format: evidence.txt, proteinGroups.txt; plus externally generated annotation file
OpenSWATH format: pyprophet export file; plus externally generated annotation file
Annotations as tabular file are needed for all input options except MSstats format
4 columns with exactly these headers: Raw.file, Condition, BioReplicate, Run; additional 5th column only for MaxQuant: IsotopeLabelType
Example file header:
Raw.file Condition BioReplicate Run IsotopeLabelType ** disease ReplA 1 L ** disease ReplA 2 L ** disease ReplB 3 L ** disease ReplB 4 L ... ... ... ... ...
Raw.file:
- OpenSWATH: File name needs to fit exactly how it is written in OpenSwatch output (e.g. "in/AA12_mzML.mzML")
- MaxQuant: File name needs to fit exactly how it is writtein in the evidence.txt "Raw file" column. (e.g. "file1.raw.thermo")
Condition: The name of the condition is not allowed to start with a number or contain any special characters
All other columns: see description above for MSstats format columns
Comparison matrix as tabular file
1st column: name of comparison
Additionally one column for each condition that is present in the tabular file. Use 1 and -1 to indicate the conditions to compare and 0 for conditions that are not compared. Multiple groups can be combined by using 0.5.
First row contains the names of the groups, they must exactly match the condition name used in the annotation file and every condition must be present, even though it will not be used for any comparison such as G4 in the example below. Order of the condition columns is irrelevant.
Each additional row represents one comparison
Example for a two group comparison
names groupA groupB groupA-groupB 1 -1Example for an experiment with 5 groups and 4 different comparisons
names G1 G2 G3 G4 G5 G2-G1 -1 1 0 0 0 G3-G5 0 0 1 0 -1 G3-G5 0 0 -1 0 1 G1+G2-G5 0.5 0.5 0 0 -1
Options
Data conversion from MaxQuant and OpenSWATH to MSstats format:
- MaxQuant input: Contaminants and reverse and only identified by site from MaxQuant tool are automatically removed during conversion
Data processing options:
Log transformation: log2 or log10 transformation of intensities
Normalization of MS runs: If there are multiple fractionations or injections for one sample, normalization is performed by each fractionation or different m/z range from multiple injections.
- equalizeMedians: The default option for normalization is equalizeMedians, where all intensities in a run are shifted by a constant, to equalize the median of intensities across runs for label-free experiment. This normalization method is appropriate when we can assume that the majority of proteins do not change across runs. Be cautious when using the equalizeMedians option for a label-free DDA dataset with only a small number of proteins. For label based experiment, equalizeMedians equalizes the median of reference intensities across runs and is generally proper even for a dataset with a small number of proteins.
- globalStandards: If you have a spiked in standard, you may set this option to define the standard with name Standardsoption.
- quantile: The distribution of all the intensities in each run will become the same across runs for label-free experiment. For label-based experiment, the distribution of all the reference intensities will become the same across runs and all the endogenous intensities are shifted by a constant corresponding to reference intensities.
- FALSE: No normalization is performed. If you had your own normalization before MSstats use this option.
Feature selection
- all: Use all features in the dataset.
- top3: Use top 3 features which have highest average of log(intensity) across runs.
- topN: Use top N (specify number) features which have highest average of log(intensity) across runs.
- highQuality: Detect and flag uninformative features (as Uninformative in the feature_quality column) and outliers (as TRUE in the is_outliercolumn). These uninformative content may be excluded from run-level summarization by setting the remove features flagged with uninformative feature quality option to TRUE.
Summarizing intensities per MS run
TMP: Tukey’s median polish. Robust parameter estimation method with median across rows and columns. Prerequisite for missing value imputation.
linear: Linear model (lmfunction). Average-based summarization.
- Account for heterogeneous variation among intensities from different features: Yes: assumes equal variance among intensities from features. No: means that we cannot assume equal variance among intensities from features, then we will account for heterogeneous variation from different features
Missing value imputation:
Impute Missing Values: Only possible for Summarization Method TMP. Censored missing values will be determined and imputed by Accelerated Failure Time model.
Remove runs which have more than 50% missing values: Yes or no.
Censored Intensity: The processing tools report missing values differently. This option is for distinguishwhich value should be considered as missing, and further whether it is censored or at random
- NA - It assumes that all NAs in Intensity column are censored.
- 0 - It assumes that all values between 0 and 1 in Intensity column are censored. If there areNAs inIntensitywith this option, NAs will be considered as random missing.
- Skyline and OpenSWATH input should use '0'. MaxQuant input should use 'NA'
Group comparison: automatic detection of differentially abundant proteins between two conditions, conditions have to be specified with the 'comparison matrix'
Quantification per sample or group: choose the corresponding output option
- Sample: relative protein abundance in each biological replicate. If there are technical replicates for biological replicates,sample quantification will be the median among technical replicates. If there is no technical replicate for biological replicate (sample), sample quantification will be the same as run-level summarization.
- Group: relative protein abundance in each condition, summarized over the biological replicates (median among sample quantification). In presence of completely missing values in a condition, the estimates will be zero
Output options
Different outputs available. Especially for studies with many proteins, it is suggested to select only the necessary pdf outputs as many of them generate one plot per protein.
MSstats log - check log file for warnings and information on the analysis steps (txt)
MSstats Rscript - can be used to re-run analysis outside Galaxy or to inspect the executed code (txt)
MSstats RawData - raw files combined into MSstats format (tabular)
MSstats FeatureLevelData - transformed, normalized, imputed intensities (tabular)
- Intensity column: includes original intensities values
- Abundance column: contains the log2 transformed and normalized intensities and it will used for run-level summarization
- Censored column: has the decision about censored missing or not, based on censored Intensity and maximum quantile for deciding censored missing values options. Abundances with TRUE value in censored column will be considered as censored missing and imputed when Missing value imputation: Yes.
MSstats ProteinLevelData - run and protein level summarized data (tabular)
- LogIntensities: log intensity summarized per run and protein, they will be used for the group comparison and summarized profile plot
- NumMeasuredFeature: shows how many features were used for summarization of the corresponding run and protein
- MissingPercentage: percentage of random and censoredmissing in the corresponding run and protein out of the total number of feature in the corresponding protein.
- more50missing: whether MissingPercentage is greater than 50% or not
- NumImputedFeatures: how many features were imputed in the corresponding run and protein
MSstats QCPlot - log2 intensity boxplot for all proteins and run on first page, followed by one boxplot per protein (pdf)
MSstats ProfilePlot - log2 intensity profiles one plot per protein and run (pdf)
- Profile plot helps identify potential sources of variation (both variation of interest and nuisance variation) for each protein: show individual measurements for each peptide (peptide for DDA, transition for SRM orDIA) across runs, grouped per condition. Each peptide has a different color/type layout. Disconnected linesshow that there are missing value (NA).
MSstats ProfilePlot_wSummarization - log2 intensity profiles one plot per protein and run with run summarization (pdf)
- Run-level summarized data per protein. The same peptides (or transition) in the first plot are presented in grey, with the summarized values overlaid in red.
MSstats ConditionPlot - log2 intensity range for each protein and condition (pdf)
- Visualizes potential systematic differences in protein intensities between conditions. Dots indicate the mean of log2 intensities for each condition, error bars indicate the confidence interval with 0.95 significant level for each condition. The intervals are for descriptive purposes only.
Sample Quantification Matrix/Long Table - relative protein abundance in each biological replicate in matrix (rows are proteins, and columns are combinations of biological replicate and group, filled with LogIntensities) or long format (row corresponding to relative protein abundances, and columns are Protein, Group, BioReplicate, LogIntensities) (tabular)
- If there are technical replicates for biological replicates, sample quantification will be the median among technical replicates. If there is no technical replicate for biological replicate (sample), sample quantification will be the same as run-level summarization. In presence of completely missing values in a biological replicate, the estimates will be zero.
Group Quantification Matrix/Long Tableuant_group_matrix - relative protein abundance in each condition in matrix (rows are proteins, and columns are groups) or long format (row corresponding to relative protein abundances, and columns are Protein, Group and LogIntensities) (tabular)
- Outputs the estimates of relative protein abundance in each condition, summarized over the biological replicates (median among sample quantification). In presence of completely missing values in a condition, the estimates will be zero.
MSstats ComparisonFittedModel (txt)
MSstats ComparisonResult - summary of statistical results per protein and comparison (tabular)
- Label: name of the comparison (e.g. condition1 - condition2)
- log2FC: log2 fold change for the given comparison name, e.g. condition1-condition2: positive values mean more abundant in condition1, negative values mean more abundant in condition2
- SE: standard error of the log2 fold change
- Tvalue: test statistic of the Student test
- DF: degree of freedom of the Student test
- pvalue: raw p-values
- adj. pvalue: adjusted p-values among all the proteins in the specific comparison
- issue: shows if there is any issue for inference in corresponding protein and comparison,for example,OneConditionMissing or CompleteMissing. If one of condition for compariosn is completely missing, it would flag with OneConditionMissing with adj.pvalue=0 and log2FC=Inf or -Inf even though pvalue=NA. For example, if you want to compare ‘condition1-condition2’, but condition2 has complete missing, log2FC=Inf and adj.pvalue=0. SE,Tvalue, and pvalue will be NA. If you want to compare ‘conditions - condition2’, but condition1 has complete missing, then log2FC=-Inf and adj.pvalue=0. But, please be careful for using this log2FC and adj.pvalue.
MSstats ModelQC - summary statistics per run and protein (tabular)
MSstats QQPlot - one QQplot per protein (pdf)
- Normal quantile-quantile plots for each protein, taking as input the results of model fitting and testing in groupComparison. Only large deviations of transition intensities from the straight line are problematic and indicate that the assumption of the normal distribution of the measurement errors may not hold.
MSstats ResiudalPlot - one residual plot per protein (pdf)
- Residual plot shows variance of the residuals that is associated with the mean feature intensity. Any specific pattern, such as increasing or decreasing by predicted abundance, is problematic and indicates that the assumption of constant variance of the measurement error may not hold.
MSstats VolcanoPlot - one volcano plot per comparison (pdf)
- Visualizes the outcome of one comparison between conditions for all the proteins, and combine the information on statistical and practical significance. The y-axis displays the FDR-adjusted p-values on the negative log10 scale, representing statistical significance. The horizontal dashed line shows the FDR cutoff. The points above the FDR cutoff line are statistically significant proteins that are differentially abundant across conditions. These points are colored in red and blue for upregulated and downregulated proteins, respectively. The x-axis is the model-based estimate of fold change on log scale and represents practical significance. It is possible to specify a practical significance cutoff based on the estimate of fold change in addition to the statistical significance cutoff. If the fold change cutoff is specified, the points above the horizontal cutoff line but within the vertical cutoff line will be considered as not differentially abundant (and will be colored in black).
MSstats Heatmap - needs at least 2 comparisons, one heatmap for all proteins and comparisons (pdf)
- Illustrates the patterns of up- and down-regulation of proteins in several comparisons. Columns in the heatmaps are comparison of conditions assigned in contrast matrix, and rows are proteins. The heatmaps display signed FDR-adjusted p-values of the tests, colored in red/blue for significantly up-/down-regulated proteins, while taking into account the specified FDR cutoff and the additional optional fold change cutoff. Brighter colors indicate stronger evidence in favor of differential abundance. Black color represents proteins that are not significantly differentially abundant.
MSstats ComparisonPlot - log2 intensity range for each protein and comparison (pdf)
- Illustrates model-based estimates of log-fold changes, and the associated uncertainty, in several comparisons of conditions for one protein. X-axis is the comparison of interest. Y-axis is the log fold change. The dots are the model-based estimates of log-fold change, and the error bars are the model-based 95% confidence intervals. For simplicity, the confidence intervals are adjusted for multiple comparisons within protein only, using the Bonferroni approach. For proteins with N comparisons, the individual confidence intervals are at the level of 1-sig/N.
For additional help please visit the MSstats documentation