Network Inference Toolbox

This defines Singularity containers to run a bunch of tools for Gene Regulatory Network inference.

Dependencies

Building the containers requires superuser privileges and Singularity installed. This was tested with v3.7, but other v3 versions should work too. GNU make is optional.

For running the containers, Singularity is required as well. (Older versions should work fine but were not tested.)

Note	Building and running the containers on Windows Subsystem for Linux is currently not supported (and does not work).

Overview

Currently the following containers are defined. They are supposed not to write to your bind mounts (like home) implicitly; if they do, that is a bug.

You can build a single container by calling

sudo singularity build ⟨container.sif⟩ ⟨container.def⟩

and should put the resulting containers into bin. Alternatively, with make installed, you can just run

sudo make

in the repository root to build all containers.

The containers also each contain a singularity app specification that writes a digest of the container definition to standard output. This can be useful to detect changes to the containers downstream because the blobs are unfortunately not unique given their dependencies.

arboreto.sif

A current Python with the arboreto package installed, which contains both a Python implementation of GENIE3 and its “GRNBOOST2” reimplementation.

The run_arboretum.py [sic] script from the arboreto repository is installed as a start script. It has its own help mechanism (through argparse); call it with --help, e.g. like:

bin/arboreto.sif --help

NetworkInference.jl.sif

An older Julia (v1.0.5) with the NetworkInference package installed and precompiled. All cache directories are frozen in the image to prevent it from polluting your bind mounts. This means however that new Julia code cannot be automatically compiled. If you need anything beyond running NetworkInference, you have the following options:

• Ensure the code is precompiled by calling it in blueprints/NetworkInference.jl/environment/precompile and then rebuilding the container.

• Set the JULIA_DEPOT_PATH environment variable to a writable location when running the container.

• Run the container in sandbox mode.

The container start script just redirects to the julia executable, so you can either just run bin/NetworkInference.jl.sif to drop into a Julia shell or pass it a script to run. Call it with --help to learn more:

bin/NetworkInference.sif --help

R-multitool.sif

This container contains a current R installation with a bunch of network inference packages preinstalled. This includes both standard R packages and code that is not prepackaged. The available tools are:

• BTR, which was removed from CRAN and is therefore installed manually, available via library(BTR).

• GENIE3 from Bioconductor, an R implemention of GENIE3 with a C kernel, available via library(GENIE3).

• GENIE3-R, an R implementation of GENIE3, available at /opt/GENIE3_R inside the container. Source /opt/GENIE3_R/GENIE3.R to use it.

• LEAP from CRAN, available via library(LEAP).

• ppcor from CRAN, available via library(ppcor).

• SCODE-R, the R version of the SCODE package, available at /opt/SCODE inside the container. Source /opt/SCODE/SCODE.R to use it.

• SINCERITIES-R, the R version of the SINCERITIES package, available at /opt/SINCERITIES. This is untested at the moment because I did not find good documentation. Source /opt/SINCERITIES/MAIN.R to use it.