wwood/aerobicity

Prediction of aerobicity (whether an bacteria or archaeon is aerobic) based on gene copy numbers. The prediction problem is posed as a 2-class problem (the prediction is either aerobic or anaerobic).

This predictor was used in this (currently pre-publication) manuscript, please cite it if appropriate:

Davín, A.A., Woodcroft, B.J., Soo, R.M., Morel, B., Murali, R., Schrempf, D., Clark, J.W., Álvarez-Carretero, S., Boussau, B., Moody, E.R. and Szánthó, L.L., 2025. A geological timescale for bacterial evolution and oxygen adaptation. Science, 388(6742), p.eadp1853. https://doi.org/10.1126/science.adp1853

Installation

First ensure you have installed git-lfs (including running git lfs install), as described at https://www.atlassian.com/git/tutorials/git-lfs#installing-git-lfs

Then clone this repository, using

git clone https://huggingface.co/wwood/aerobicity
git lfs fetch --all
git lfs pull

Then setup the conda environment:

cd aerobicity
mamba env create -p env -f env-apply.yml
conda activate ./env

and download the eggNOG database. We use version 2.1.3, as specified in the env-apply.yml conda environment file, because this is what the predictor was trained on. The eggNOG database is large, so it is not included in the repository. To download it, run:

mkdir eggNOG
download_eggnog_data.py --data_dir ./eggNOG

Usage

To apply the predictor, run against a test genome:

./17_apply_to_proteome.py --protein-fasta data/RS_GCF_000515355.1_protein.faa --eggnog-data-dir eggNOG/ 
--models XGBoost.model --output-predictions predictions.csv

The predictions are then in predictions.csv. In the predictions output file, a prediction of 0 corresponds to a anaerobic prediction, and 1 corresponds to an aerobic prediction.

To run on your genomes, provide its protein fasta (i.e. the result of running prodigal on it), and use that instead of data/RS_GCF_000515355.1_protein.faa in the above command.