This repository contains the R code used to generate all the results in the article:
Picart-Armada, S., Thompson, W. K., Buil, A., & Perera-Lluna, A. (2020). The effect of statistical normalisation on network propagation scores. BioRxiv.
The main purpose of the article is to understand in a quantitative manner:
- The presence of bias in classical diffusion scores, its classification into expected value- and variance-related, and its quantification in three use cases.
- The connections of the exact expected values and variances of the null diffusion scores (i.e. those arising from a permuted input) to the properties of the nodes (positives, negatives) and the network (spectral graph theory).
- Exact equivalences between various diffusion scores and other mathematical properties.
- The impact and the interpretation of normalising diffusion scores using exact z-scores.
The calculations for classical and normalised propagation scores were carried out using the companion package diffuStats, available in Bioconductor and citable as:
Picart-Armada, S., Thompson, W. K., Buil, A., & Perera-Lluna, A. (2018). diffuStats: an R package to compute diffusion-based scores on biological networks. Bioinformatics, 34(3), 533-534.
Overall, there are four main analyses, which are aligned with the folder structure.
| Analysis | Purpose | Folder |
|---|---|---|
| Mathematical properties | Describe the statistics | 00_properties |
| Synthetic signals on yeast interactome | Understand biased and unbiased true signals | 01_synthetic |
| Simulated gene expression in a human interactome | Influence of statistical background to the performance | 02_dlbcl |
| Prospective pathway gene prediction | Variance-related bias use case | 03_retroData |
In general, the metadata directories contain the output of sessionInfo(), whereas params.R contain configuration settings and general parameters and aux_*.R contain auxiliary functions with the purpose specified in the file name.
The directories that match the numeric prefix of a report (e.g. the report 02_dlbcl/01_analysis.Rmd and the directory 02_dlbcl/01_network) typicall contain its output files.
Reports may include bibliographical entries (*.bib).
diffuBench
├── 00_metadata
├── 00_packages: source code for custom packages
│ ├── diffuStats_0.99.9.tar.gz older version of diffuStats (newest in Bioconductor)
│ └── retroData_0.0.2.tar.gz data package
├── 00_networkCuration: study of the impact of filtering edges by confidence
│ ├── 10_bibliography.bib
│ └── 10_supplement.Rmd: supplementary material report
├── 00_properties: study of the mathematical properties
│ ├── 00_metadata
│ ├── 01_equivalences.Rmd: explore equivalences between scores
│ ├── 02_spectral_properties.Rmd: plot spectral properties of null covariance
│ └── params.R
├── 01_synthetic: case study of synthetic signals on a yeast network
│ ├── 10_bibliography.bib
│ ├── 10_synth_yeast.Rmd: supplementary material report
│ ├── aux_generate_input.R
│ └── params.R
├── 02_dlbcl: case study of synthetic gene expression in a human interactome
│ ├── 00_metadata
│ ├── 01_analysis.Rmd: prepare interactome and KEGG data
│ ├── 01_kegg: tables from the KEGG database
│ ├── 02_kernel.Rmd: compute graph kernel
│ ├── 03_generateInputs.Rmd: compute synthetic differential gene expression
│ ├── 04_computeScores.Rmd: compute propagation scores and fit regression models
│ ├── 05_plots.Rmd: plot the metrics
│ ├── 06_positive_analysis.Rmd: study the properties of the pathway nodes (positives)
│ ├── 10_bibliography.bib
│ ├── 10_main: figures for the main body
│ ├── 10_supplement.Rmd: supplementary file report
│ ├── aux_deepestdir.R
│ ├── aux_param2name.R
│ ├── aux_sample_genes.R
│ └── params.R
├── 03_retroData: case study of prospective pathway prediction
│ ├── 00_metadata
│ ├── 01_descriptive.Rmd: data preprocessing and description
│ ├── 02_diffusion_scores.Rmd: computation of diffusion scores
│ ├── 03_statistics.Rmd: metrics and rankings
│ ├── 04_plots.Rmd: regression models and plotting
│ ├── 10_bibliography.bib
│ ├── 10_main: figures for the main body
│ ├── 10_supplement.Rmd: supplementary file report
│ ├── aux_compute_partitions.R
│ └── params.R
├── 04_networkCuration: study of the filtering effect in the null models
│ └── 01_descriptive_biogrid.Rmd: generation of kernels for varying filterings
├── 0a_network_choice.Rmd: complementary files to choose the synthetic network models
├── .gitignore: exclude some filetypes from version control
├── helper_funs.R: functions for small calculations and plotting
├── LICENSE: terms for code reuse and distribution
├── Makefile: basic recipes to regenerate the analysis and the reports
└── packrat: internal files for package management (including package source code)
The whole analysis was run on the following desktop hardware:
- Intel(R) Core(TM) i5 CPU 650 @ 3.20GHz (4 threads)
- 16GB RAM
And under the following operating system/environment:
- R version 3.5.3 (2019-03-11)
- Platform: x86_64-pc-linux-gnu (64-bit)
- Running under: Ubuntu 16.04.6 LTS
The packrat R package was used to record the working versions of the packages hereby used.
Using packrat::on() should start the process of either installing them, or adding them to .libPaths().
Run packrat::packify() to generate a proper .Rprofile file that automates turning packrat on for clean sessions.
git clone https://github.com/b2slab/diffuBench
cd diffuBenchIf this is the first time you run the analysis, you need to install the required packages. To that end, first run
packrat::packify() Then trigger the (one-off) installation of the packages, this will take long as there are around 200 of them
packrat::on() The packages will be installed into ./packrat, without interfering with your R system libraries.
For this purpose, you can make use of the Makefile recipes.
Note that the first lines, displayed below, may require modification in your system:
# choose R version (I forced 3.5.3)
RSCRIPT := /opt/R/3.5.3/bin/Rscript
LATEX_CMD = pdflatex -synctex=1 -interaction=nonstopmode --shell-escape
BIB_CMD = bibtexAlthough it is recommended to use R 3.5.3, the code was also run on 3.4.4, 3.5.1 and 3.5.2.
You might need to change the path to the R executable in your system.
The params.R files contain paths that you might want to change, typically to store large kernel matrices.
E.g. file_kernel <- "~/big/devel/big/diffusion/kernell_dlbcl.RData" in 02_dlbcl/params.R.
You might want to change those to somewhere convenient in your machine.
Now simply run
make props demos filters s1 s2 s3This would take up to one day to rebuild everything on our machine.
You can direct your complaints at sergi.picart at upc dot edu.