In const-ae/proDA: Differential Abundance Analysis of Label-Free Mass Spectrometry Data

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Introduction

After running your samples on a mass spectrometer, you want to find out if there are interesting patterns in the data. But the first challenge is how do you get the data from the files that your mass spectrometer produced into R?

In the following, I will describe several ways of importing data from MaxQuant. The general approaches will also be applicable to data from other tools, you will just have to adapt the column names.

MaxQuant File Overview

MaxQuant is a popular tool for identifying, integrating, and combining MS peaks to derive peptide and protein intensities. MaxQuant produces several output files including proteinGroups.txt. It is usually a tab separated table with a lot of different columns, which can make it difficult to not get overwhelmed with information.

The most important columns that every proteinGroups.txt file contains are

• Protein IDs: a semicolon delimited text listing all protein identifiers that match an identified set of peptides. Most of the time this is just a single protein, but sometimes proteins are so similar to each other because of gene duplication that it was not possible to distinguish them.

• Majority protein IDs: a semicolon delimited text that lists all proteins from the Protein IDs column which had more than half of their peptides identified.

• Identification type [SAMPLENAME]: For each sample there is one column that explains how the peptide peaks where identified. Either they were directly sequenced by the MS2 ("By MS/MS") or by matching the m/z peak and elution timing across samples ("By matching").

• Intensity [SAMPLENAME]: The combined intensity of the peptides of the protein. Missing or non-identified proteins are simply stored as 0. In a label-free experiment, this is also often called LFQ Intensity [SAMPLENAME].

• iBAQ [SAMPLENAME]: iBAQ is short for intensity-based absolute quantification. It is an attempt to make intensity values comparable across proteins. Usually the intensity values are only relative, which means that they are only comparable within one protein. This is because differences in ionization and detection efficiency. It is usually better to just compare the Intensity columns to identify differentially abundant proteins.

• Only identified by site: Contains a "+" if the protein was only identified by a modification site.

• Reverse: Contains a "+" if the protein matches the reversed part of the decoy database.

• Contaminant: Contains a "+" if the protein is a commonly occurring contaminant.

The last three columns are commonly used to filter out false positive hits.

The full information what each column means is provided in the tables.pdf file in the MaxQuant output folder.

Workflow

Our goal is to turn this complicated table into a useable matrix or a SummarizedExperiment object. There are several ways to achieve this:

1. Use the base R functions (read.delim() and [<-) to read in the data
2. Use the tidyverse packages to load the file and turn it into a useable object
3. Use the DEP package and the import_MaxQuant() function

I will demonstrate each approach using an example file that comes with this package.

The example file contains the LFQ data from a BioID experiment in Drosophila melanogaster. 11 different Palmitoyltransferases (short DHHC) were tagged with a promiscuous biotin ligase and all biotinylated proteins were enriched and identified using label-free mass spectrometry. The conditions are named after the tagged DHHC and the negative control condition is called S2R for the cell line. Each condition was measured in triplicates, which means that there are a total of 36 samples To make the file smaller, I provide a reduced data set which only contains the first 122 rows of the data.

Base R

The example file is located in

system.file("extdata/proteinGroups.txt", 
            package = "proDA", mustWork = TRUE)

In this specific file, all spaces have been replaced with dots. This is an example how each output file from MaxQuant slightly differs. This can make it difficult to write a generic import function. Instead I will first demonstrate the most general approach which is to simply use the base R tools for loading the data and turning it into useful objects.

The first step is to load the full table.

full_data <- read.delim(
    system.file("extdata/proteinGroups.txt", 
                package = "proDA", mustWork = TRUE),
    stringsAsFactors = FALSE
)

head(colnames(full_data))

Next, I create a matrix of the intensity data, where each sample is a column and each protein group is a row.

# I use a regular expression (regex) to select the intensity columns
intensity_colnames <- grep("^LFQ\\.intensity\\.", colnames(full_data), value=TRUE)

# Create matrix which only contains the intensity columns
data <- as.matrix(full_data[, intensity_colnames])
colnames(data) <- sub("^LFQ\\.intensity\\.", "", intensity_colnames)
# Code missing values explicitly as NA
data[data == 0] <- NA
# log transformation to account for mean-variance relation
data <- log2(data)
# Overview of data
data[1:7, 1:6]
# Set rownames after showing data, because they are so long
rownames(data) <- full_data$Protein.IDs

In the next step I will create an annotation data.frame that contains information on the sample name, the condition and the replicate.

annotation_df <- data.frame(
    Condition = sub("\\.\\d+", "", sub("^LFQ\\.intensity\\.", 
                                       "", intensity_colnames)),
    Replicate = as.numeric(sub("^LFQ\\.intensity\\.[[:alnum:]]+\\.", 
                               "", intensity_colnames)),
    stringsAsFactors = FALSE, row.names = colnames(data)
)
head(annotation_df)

We can use this data to fit the probabilistic dropout model and test for differentially abundant proteins.

# Not Run
library(proDA)
fit <- proDA(data, design= annotation_df$Condition, col_data = annotation_df)
test_diff(fit, contrast = CG1407 - S2R)
# End Not Run

Optionally, we can turn the data also into a SummarizedExperiment or MSnSet object

library(SummarizedExperiment)
library(MSnbase)

library(SummarizedExperiment)
se <- SummarizedExperiment(SimpleList(LFQ=data), colData=annotation_df)
rowData(se) <- full_data[, c("Only.identified.by.site", 
                             "Reverse", "Potential.contaminant")]
se

library(MSnbase)

fData <- AnnotatedDataFrame(full_data[, c("Only.identified.by.site", 
                                 "Reverse", "Potential.contaminant")])
rownames(fData) <- rownames(data)
ms <- MSnSet(data, pData=AnnotatedDataFrame(annotation_df), fData=fData)
ms

Both input types are also accepted by proDA.

# Not Run
library(proDA)
fit <- proDA(se, design = ~ Condition - 1)
test_diff(fit, contrast = ConditionCG1407 - ConditionS2R)
# End Not Run

Tidyverse

The tidyverse is a set of coherent R packages that provide many useful functions for common data analysis tasks. It replicates many of the functionalities already available in base R packages, but learns from its mistakes and avoids some of the surprising behaviors. For example strings are never automatically converted to factors. Another popular feature in the tidyverse is the pipe operator (%>%) that makes it easy to chain complex transformations.

library(dplyr)
library(stringr)
library(readr)
library(tidyr)
library(tibble)

library(dplyr)
library(stringr)
library(readr)
library(tidyr)
library(tibble)
# Or short 
# library(tidyverse)

I first load the full data file

# The read_tsv function works faster and more reliable than read.delim
# But it sometimes needs help to identify the right type for each column,
# because it looks only at the first 1,000 elements. 
# Here, I explicitly define the `Reverse` column as a character column
full_data <- read_tsv(
    system.file("extdata/proteinGroups.txt", 
                package = "proDA", mustWork = TRUE),
    col_types = cols(Reverse = col_character())
)

full_data

Next, I create a tidy version of the data set. I pipe (%>%) the results from each transformation to the next transformation, to first select the columns of interest, reshape (gather) the dataset from wide to long format, and lastly create new columns with mutate.

# I explicitly call `dplyr::select()` because there is a naming conflict
# between the tidyverse and BioConductor packages for `select()` function
tidy_data <- full_data %>%
    dplyr::select(ProteinID=Protein.IDs, starts_with("LFQ.intensity.")) %>%
    gather(Sample, Intensity, starts_with("LFQ.intensity.")) %>%
    mutate(Condition = str_match(Sample, 
                 "LFQ\\.intensity\\.([[:alnum:]]+)\\.\\d+")[,2]) %>%
    mutate(Replicate = as.numeric(str_match(Sample, 
                 "LFQ\\.intensity\\.[[:alnum:]]+\\.(\\d+)")[,2])) %>%
    mutate(SampleName = paste0(Condition, ".", Replicate))

tidy_data

Using the tidy data, I create the annotation data frame and the data matrix.

data <- tidy_data %>%
    mutate(Intensity = ifelse(Intensity == 0, NA, log2(Intensity))) %>%
    dplyr::select(ProteinID, SampleName, Intensity) %>%
    spread(SampleName, Intensity) %>%
    column_to_rownames("ProteinID") %>%
    as.matrix()

data[1:4, 1:7]

annotation_df <- tidy_data %>%
    dplyr::select(SampleName, Condition, Replicate) %>%
    distinct() %>%
    arrange(Condition, Replicate) %>%
    as.data.frame() %>%
    column_to_rownames("SampleName")

annotation_df

Optionally, we can again turn this into a SummarizedExperiment or MSnSet object

library(SummarizedExperiment)
se <- SummarizedExperiment(SimpleList(LFQ=data), colData=annotation_df)
rowData(se) <- full_data[, c("Only.identified.by.site", 
                             "Reverse", "Potential.contaminant")]
se

library(MSnbase)

fData <- AnnotatedDataFrame(full_data[, c("Only.identified.by.site", 
                                 "Reverse", "Potential.contaminant")])
rownames(fData) <- rownames(data)
ms <- MSnSet(data, pData=AnnotatedDataFrame(annotation_df), fData=fData)
ms

Both input types are also accepted by proDA.

# Not Run
library(proDA)
fit <- proDA(se, design = ~ Condition - 1)
test_diff(fit, contrast = ConditionCG1407 - ConditionS2R)
# End Not Run

DEP

DEP is a BioConductor package that is designed for the analysis of mass spectrometry data. It provides helper functions to impute missing values and makes it easy to run limma on the completed dataset.

To load the data, we need to provide all the column names of the intensity values. I then call the import_MaxQuant() function that directly creates a SummarizedExperiment object.

library(DEP)

full_data <- read.delim(
    system.file("extdata/proteinGroups.txt", 
                package = "proDA", mustWork = TRUE),
    stringsAsFactors = FALSE
)


exp_design <- data.frame(
   label =c("LFQ.intensity.CG1407.01", "LFQ.intensity.CG1407.02",  "LFQ.intensity.CG1407.03",  "LFQ.intensity.CG4676.01",  "LFQ.intensity.CG4676.02", "LFQ.intensity.CG4676.03", "LFQ.intensity.CG51963.01", "LFQ.intensity.CG51963.02", "LFQ.intensity.CG51963.03","LFQ.intensity.CG5620A.01", "LFQ.intensity.CG5620A.02", "LFQ.intensity.CG5620A.03", "LFQ.intensity.CG5620B.01","LFQ.intensity.CG5620B.02", "LFQ.intensity.CG5620B.03", "LFQ.intensity.CG5880.01", "LFQ.intensity.CG5880.02",  "LFQ.intensity.CG5880.03",  "LFQ.intensity.CG6017.01",  "LFQ.intensity.CG6017.02", "LFQ.intensity.CG6017.03", "LFQ.intensity.CG6618.01",  "LFQ.intensity.CG6618.02",  "LFQ.intensity.CG6618.03",  "LFQ.intensity.CG6627.01", "LFQ.intensity.CG6627.02", "LFQ.intensity.CG6627.03",  "LFQ.intensity.CG8314.01", "LFQ.intensity.CG8314.02",  "LFQ.intensity.CG8314.03", "LFQ.intensity.GsbPI.001", "LFQ.intensity.GsbPI.002",  "LFQ.intensity.GsbPI.003",  "LFQ.intensity.S2R.01", "LFQ.intensity.S2R.02", "LFQ.intensity.S2R.03"),
   condition = c("CG1407", "CG1407", "CG1407", "CG4676", "CG4676", "CG4676", "CG51963", "CG51963", "CG51963", "CG5620A", "CG5620A", "CG5620A", "CG5620B", "CG5620B", "CG5620B", "CG5880", "CG5880", "CG5880", "CG6017", "CG6017", "CG6017", "CG6618", "CG6618", "CG6618", "CG6627", "CG6627", "CG6627", "CG8314", "CG8314", "CG8314", "GsbPI", "GsbPI", "GsbPI", "S2R", "S2R", "S2R" ),
   replicate = rep(1:3, times=12),
   stringsAsFactors = FALSE
)

se <- import_MaxQuant(full_data, exp_design)
se
assay(se)[1:5, 1:5]

Again, we can run proDA on the result:

# Not Run
library(proDA)
fit <- proDA(se, design = ~ condition - 1)
# Here, we need to be specific, because DEP also has a test_diff method
proDA::test_diff(fit, contrast = conditionCG1407 - conditionS2R)
# End Not Run

Session Info

sessionInfo()

const-ae/proDA documentation built on Oct. 31, 2023, 9:39 p.m.