Abstract: Despite the importance of microbial dysbiosis in human disease, the phenomenon remains poorly understood. We provide the first comprehensive and predictive model of dysbiosis at ecosystem-scale, leveraging our new machine learning method for efficiently inferring compact and interpretable dynamical systems models. Coupling this approach with the most densely temporally sampled interventional study of the microbiome to date, using microbiota from healthy and dysbiotic human donors that we transplanted into mice subjected to antibiotic and dietary interventions, we demonstrate superior predictive performance of our method over state-of-the-art techniques. Moreover, we demonstrate that our approach uncovers intrinsic dynamical properties of dysbiosis driven by destabilizing competitive cycles, in contrast to stabilizing interaction chains in the healthy microbiome, which have implications for restoration of the microbiome to treat disease.
Important links
- Main Paper (Pre-print): "Intrinsic instability of the dysbiotic microbiome revealed through dynamical systems inference at scale"
- Associated GitHub repo for the ML model: "MDSINE2"
- Folder containing tutorials as notebooks exploring the model, data and paper that can be opened directly in Google Colab
- Raw sequences from longitudinal experiments on NCBI
Pre-print
@article {Gibson2021.12.14.469105,
author = {Gibson, Travis E and Kim, Younhun and Acharya, Sawal and Kaplan, David E and DiBenedetto, Nicholas and Lavin, Richard and Berger, Bonnie and Allegretti, Jessica R and Bry, Lynn and Gerber, Georg K},
title = {Intrinsic instability of the dysbiotic microbiome revealed through dynamical systems inference at scale},
year = {2021},
doi = {10.1101/2021.12.14.469105},
publisher = {Cold Spring Harbor Laboratory},
URL = {https://www.biorxiv.org/content/early/2021/12/16/2021.12.14.469105},
journal = {bioRxiv}}We have provided interactive notebooks for parsing of raw and pre-preprocessed data, performing inference with our model, and reproducing the manuscript figures from pickle files containing posterior samples from full inference runs. Simply go to google_colab/ and the notebooks can be launched directly into Google Colab
This section outlines how to run MDSINE2 analysis on our dataset in full, with bash, conda and git.
One must first install the MDSINE2 package, according to the following instructions. The recommended setup starts out by creating a new conda environment. The package was developed and tested on python 3.7.3.
conda create -n mdsine2 -c conda-forge python=3.7.3
conda activate mdsine2
Next, clone and install the core MDSINE2 package (MCMC implementation) from the package repository (https://github.com/gerberlab/MDSINE2).
git clone https://github.com/gerberlab/MDSINE2
pip install MDSINE2/.
Next, clone this repository which contains the data and scripts to perform the analysis.
git clone https://github.com/gerberlab/MDSINE2_Paper
cd MDSINE2_Paper
Once the above installation done, one can run a local copy of the jupyter notebooks found in google_colab/.
conda install -c conda-forge jupyterlab
jupyter-notebook
Navigate to google_colab/ to access the notebooks.
The inference performed in the jupyter notebooks are miniature versions (so as to execute in a reasonable amount of time). For the full run, assuming that the MDSINE2 core package is installed, follow the instructions located in the analysis subfolder.