This repository contains the code used to generate the results and generate the manuscript published in:
Vandenbulcke, S., Vanderaa, C., Crook, O., Martens, L. & Clement, L. msqrob2TMT: robust linear mixed models for inferring differential abundant proteins in labelled experiments with arbitrarily complex design. bioRxiv 2024.03.29.587218 (2024) doi:10.1101/2024.03.29.587218
We also provide 2 tutorial vignettes that illustrate how to apply msqrob2TMT in practice.
Labelling strategies in mass spectrometry (MS)-based proteomics enhance sample throughput by enabling the acquisition of multiplexed samples within a single run. However, contemporary experiments often involve increasingly complex designs, where the number of samples exceeds the capacity of a single run, resulting in a complex correlation structure that must be addressed for accurate statistical inference and reliable biomarker discovery. To this end, we introduce msqrob2TMT, a suite of mixed model-based workflows specifically designed for differential abundance analysis in labelled MS-based proteomics data. msqrob2TMT accommodates both sample-specific and feature-specific (e.g., peptide or protein) covariates, facilitating inference in experiments with arbitrarily complex designs and allowing for explicit correction of feature-specific covariates. We benchmark our innovative workflows against state-of-the-art tools, including DEqMS, MSstatsTMT, and msTrawler, using two spike-in studies. Our findings demonstrate that msqrob2TMT offers greater flexibility, improved modularity, and enhanced performance, particularly through the application of robust ridge regression. Finally, we demonstrate the practical relevance of msqrob2TMT in a real mouse study, highlighting its capacity to effectively account for the complex correlation structure in the data.
To demonstrate the performance and application of msqrob2TMT, we used 3 datasets.
The spikein1 dataset has been published by Huang et al. 2020. It has the following design: UPS1 peptides at concentrations of 500, 333, 250, and 62.5 fmol were spiked into 50 µg of SILAC HeLa peptides. This series forms a dilution gradient of 1, 0.667, 0.5, and 0.125 relative to the highest UPS1 peptide concentration (500 fmol). A reference sample was prepared by combining the diluted UPS1 peptide samples with 50 µg of SILAC HeLa peptides. Each dilution and the reference sample were processed in duplicate, yielding a total of ten samples. These samples were subsequently labelled using TMT10-plex reagents and combined for LC-MS/MS analysis, hereafter referred to as a "mixture." This protocol was repeated five times. Each mixture was analysed in technical triplicate, resulting in a total of 15 MS runs. This experimental design simulates a scenario with 10 biological replicates per condition, consisting of 2 replicates per condition within each of the five mixtures.
The spikein1 dataset has been published by O'Brien et al. 2024. It is an experiment which consists of a constant mouse plasma background in which yeast proteins were then spike-in with the following dilution ratios 1, 2, 3, 5, 11, 14, 18, 20, 24, 28 and 32. The yeast proteins were harvested from different media; glycerol + ammonium sulfate, galactose + urea, galactose + monosodium glutamate, galactose + ammonium sulfate, glucose + ammonium sulfate, glucose + monosodium glutamate, and glucose + urea. The samples were labelled using the TMTpro 18-plex and pooled in six mixtures, each containing 15 samples. Two of the remaining TMT channels are reference channels, one reference channel is the combination of one batch and the other reference channel consists of all batches. For this dataset the ground truth is known: a true positive is a yeast protein that is returned as significant while a significant mouse protein is a false positive.
The mouse dataset has been published by Plubell et al. 2017. Twenty mice were divided into groups to investigate the effects of low-fat (LF) and high-fat (HF) diets. The experiment involves two factors: the type of diet (LF or HF), and the duration of the diet, categorised as either short-term (8 weeks) or long-term (18 weeks). This design resulted in four distinct groups corresponding to each combination of diet type and duration. Five mice, i.e. biological replicates, were randomly assigned to each group and samples from their epididymal adipose tissue were randomly assigned to three TMT10-plex mixtures. Within each mixture, two reference channels were included, each containing pooled samples representing a range of peptides from all samples. Not all TMT channels were utilised, leading to an unbalanced experimental design. Finally, each TMT mixture was fractionated into nine parts and analysed using synchronous precursor selection, culminating in a total of 27 MS runs.
The repository contains 4 main folders:
To illustrate how to use msqrob2TMT, we provide 2 tutorial vignettes:
spikein1.qmd: the vignette presents the theoretical aspects and the practical implementation of labelling-based MS proteomics data analysis side-by-side. As this tutorial focuses on a spike-in dataset, the impact of modelling decision on the accuracy of the method are illustrated.mouse.qmd: the vignette illustrates the application of msqrob2TMT on a real-life dataset, showing how to translate biological questions into statistical model definition.
This folder contains all the code used to preprocess the data, run the different modelling methods and to generate the figures from the manuscript. Scripts have been organised by dataset, one folder for each dataset.
This folder contains all the figures generated by the scripts and included in the manuscript.
This folder will contain intermediate tables generated by the scripts.
This folder is empty on GitHub, but can be populated with the files in
the processed.zip archive provided on our Zenodo
repository.
The data from the MSstatsTMT spike-in study, the msTrawler multibatch benchmarking study and the mouse case study are available in the PRIDE repositories under PXD0015258, PXD036799, and PXD005953, respectively. For the reproducibility of results, the data for PXD0015258, PXD036799, and PXD005953 were downloaded from MassIVE (RMSV000000265), Google Cloud storage, and MassIVE (RMSV000000264), respectively, and timestamped by creating a Zenodo repository.
All input data are automatically downloaded within the scripts, using
the Bioconductor BiocFileCache package. This package will ensure an
efficient file management, downloading the data only once without the
need for user intervention.
There are two strategies for obtaining the computational environment
(i.e. the set of installed software) to reproduce our findings: using
Docker (recommended) or going for a manual setup. For both strategies,
you must first clone this GitHub repository on your machine either by
downloading the code as a zip file or through git.
You can reproduce the paper's figures on your local machine using
Docker. You must first install Docker, then pull the image from the
DockerHub
repository.
docker pull cvanderaa/msqrob2tmt
Then, you can start an Rstudio session within a Docker container using:
docker run \
-e PASSWORD=bioc \
-p 8787:8787 \
-v `pwd`:/home/rstudio/ \
cvanderaa/msqrob2tmt
Note you should use %CD% instead of pwd when using Windows.
Open your browser and go to http://localhost:8787. The USER is
rstudio and the password is bioc. See the DockerHub
repository
for more detailed information on getting started with Docker.
Similarly, you can run the Docker environment from the command line, providing more flexibility for developers and data analysts.
docker run -v `pwd`:/home/rstudio/ -it cvanderaa/msqrob2tmt bash
You can replace bash by your favourite shell program or by R
to immediately start an R environment.
If new dependencies are required, update the Dockerfile accordingly.
Then build the image (make sure to cd in the repo):
docker build -t cvanderaa/msqrob2tmt .
See the DockerHub repo for more information. When complete, push the new image to DockerHub:
docker push cvanderaa/msqrob2tmt
Alternatively, you can set the software environment yourself. To reproduce the article and its results, you'll need the following prerequisites:
- Install
R(suggest R version >= 4.4.0) - Install Bioconductor using
install.packages("BiocManager")Make sure that BiocManager::version() returns at least 3.20.
- Install Quarto
- Install the following R packages, make sure
msqrob2is version >= 1.10:
BiocManager::install(c("BiocParallel", "BiocStyle", "Biostrings", "curl", "DEqMS", "dplyr", "GOfuncR", "QFeatures", "lme4", "msqrob2", "MSstatsTMT", "Mus.musculus", "pryr", "quarto", "readr", "stringr", "tidyverse", "cowplot", "ggVennDiagram", "ggridges", "gt", "flextable", "ggh4x", "gprofiler2", "UpSetR", "ggborderline", "viridisLite", "impute", "ExploreModelMatrix"))Make sure that msqrob2 is at least version 1.14.1 and QFeatures is
at least version 1.16.1. If this is not the case, you can install the
latest version using
BiocManager::install(c(
"rformassspectrometry/QFeatures",
"statsOmics/msqrob2"
))Install msTrawler time stamped at commit 078eb79.
BiocManager::install("calico/msTrawler", ref = "078eb79")You should be ready to reproduce the results, although this approach can be error-prone unlike the Docker approaches.
Once you have setup the computational environment, you can run our tutorial vignettes for the spikin1 and the mouse dataset.
Once you have setup the computational environment, you can also rerun
our analyses. Each dataset can be analysed independently using the
scripts from the corresponding subfolder inscripts/. The scripts
will generate intermediate data files stored in data/ that serve as
input for following scripts. Therefore, scripts should be run in the
order shown below. However, the intermediate files stored in data/
are also provided on our Zenodo
repository. If you download the
processed.zip and unzip it in data/, the order no longer matters.
- spikein1_preprocess.R
- spikein1_deqms.R
- spikein1_msqrob2tmt.R
- spikein1_msstatstmt.R
- spikein1_mstrawler.R
- spikein1_compare_preprocessing.R
- spikein1_figures.R
- spikein2_preprocess.R
- spikein2_deqms.R
- spikein2_msqrob2tmt.R
- spikein2_msstatstmt.R
- spikein2_mstrawler.R
- spikein2_figures.R
- mouse_msqrob2tmt_mixture.R
- mouse_msqrob2tmt.R
- mouse_msstatstmt.R
- mouse_figures.R
Note that the *_figures.R scripts will generate figures in the
figs/ folder.
Vandenbulcke, S., Vanderaa, C., Crook, O., Martens, L. & Clement, L. msqrob2TMT: robust linear mixed models for inferring differential abundant proteins in labelled experiments with arbitrarily complex design. bioRxiv 2024.03.29.587218 (2024) doi:10.1101/2024.03.29.587218