ReMIND: Reconstruction of Microbial Interaction Networks using Decomposed in silico minimal exchanges

This repository contains the workflow to reconstruct metabolic interaction networks in microbial communities using decomposed in silico Minimal exchanges

More information can be found in the The package is available for both python and matlab. The results in the preprint are generated with the python version. The details on the matlab version and setup can be found in this file The package is developed using python 3.6 and run in Docker (20.10.21) containers. Tested with solvers cplex (v12.8.0) and gurobi(v9.1.2) and tested on both Linux and macOS operating systems.

Recommended to be run in docker containers, with dedicated solver installed. Installing cplex and Setting up the python API of cplex in case docker based installation is not used

Generated data used in the manuscript is available under data subfolder under hierarchical data format.

Requirements

You will need to have Git-LFS in order to properly download some binary files:

git clone https://github.com/EPFL-LCSB/remind.git /path/to/remind
cd /path/to/remind
git lfs install
git lfs pull

Further the following pip-python packages are required (can be found in detail in requirements.txt

• optlang==1.6.0
• cobra==0.17.1
• numpy<=1.17.4
• pandas
• matplotlib
• tables
• sklearn
• ipython
• jedi==0.17.2
• tqdm
• scipy
• holoviews
• matplotlib_venn
• pytfa

Container-based install

You might want to use this program inside of a container. The docker/ subfolder has all the necessary information and source files to set it up. Please clone the pytfa repository under the docker folder, or change the entrypoint and pip install it.

cd remind/python/remind/docker
./build.sh
./run.sh

Building the docker image takes approximately 5 mins.

Setup

If container-based installation is not preferred you can also install this module from source using pip, or alternatively using the environment.yml file and conda (See below) For Python 3, you might have to use pip3 instead of pip

git clone https://github.com/EPFL-LCSB/remind.git /path/to/remind/python
pip install -e /path/to/remind

The installation process should not exceed a minute if the requirements are installed. If they are not, it might take longer as the installer installs them first. Please make sure you are installing it on your virtual environment activated with python 3.6. In case you run into an issue with swiglpk package (which is required by optlang) possible workaround is to install it via conda-forge.

conda install conda-forge::swiglpk

Alternatively, we provide the environment.yml file, without cplex (as you need the license for this). Using this yml file and conda you can build the environment with all dependencies to run simulations.

conda env create -f environment.yml -n remind_test
conda activate remind_test

The installation process should not exceed a minute if the requirements are installed. If not it should take approximately 5 minutes. Please then install your solver of preference (e.g. cplex) to your environment as described above.

Quick start - Demo

As a demo following examples can be run for the 2-member honeybee gut community after building the environment or inside the docker This tutorial aims to show the step to use our framework. Can be adapted to any community with the desired extracellular environment. As mentioned above in these scripts most data is saved as .h5 binary files. For this you will need hdf files downloaded if you are running inside docker. You can find the instructions in instructions_hdf5.txt file. Or change the storing in the scripts get_dimes_tutorial.py build_community_model_from_dimes_tutorial.py and run_ilp_tutorial_community_model.py from "to_hdf" to another format (e.g. csv) "to_csv".

cd /
cd remind/projects/tutorial/

First get the DiMEs for both members by running the following bash script. Number of alternatives are limited to 10 for tutorial purposes can be changed inside the get_dimes_tutorial.py script. script get_dimes_tutorial.py by modifying the max_alternative. The time to generate alternatives for each yield regime should take around 1-2 minutes for 10 alternatives. In total running the bash file for 10 biomass yield regimes for two models should take around 20 minutes. You can comment in the bash file to generate only for certain biomass yields.

./bash_tutorial_dimes.sh

After generating the DiMEs merge the DiMEs and build the community model and save it with the following script inside Ipython. Generating the community model should take around 2-5 minutes.

ipython
run build_community_model_from_dimes_tutorial.py

The next step is to use the built community model and reconstruct the interaction networks with a user defined objective function via the ILP formulation. For this you can refer to the run_ilp_tutorial_community_model.py script. for various objective functions. To run for the indicated objective functions run the following bash script, which runs for three different objective functions, and generates alternative patters based on different integer cuts, which is commented in the bash file. For each run the simulation takes around 2-10 minutes depending on the objective funciton (e.g. depending on the number of alternatives to exhaust)

./bash_tutorial_ilp.sh

After running the ILP for various objective functions you can analyse the data inside Ipython, this step should take less than a minute.

ipython
run analysis_ilp_solutions_tutorial.py
#check the alternative maximal cooperation patterns
print(frame_int_coop.pos_int)
#check the alternative minimal competition patterns
print(frame_int_comp.neg_int)
#check the alternative minimal environments
print(frame_int_abiot.abiotic)

Reproduction of Results

To then generate the figures in the manuscript you can check the scripts inside the figures subfolder.

License

The software in this repository is put under an APACHE licensing scheme - please see the LICENSE file for more details.