Bacterial survival and morphological heterogeneity

This repository contains tables and jupyter notebooks needed to reproduce analysis and generate all plots from main and supplementary figures from the article "Extending digital biology: bacterial survival and morphological heterogeneity under antibiotic stress" by Maikranz et al. For full functionalty there is a zenodo archive with imaging files, a sql database, and neural network weights associated. Furthermore, we provide a docker container that allows to run a flask webserver which easly allows to visualise images and the associated labels. This also provides access to the labelling interface that allows context aware and context free annotations. Lastly, several scripts that produce intermediate tables depend on the webserverer running. If so this is mentioned inside the scripts.

External Data

Because of the size of the imaging files,the sql database, and neural networks weights these are provided via zenodo https://doi.org/10.5281/zenodo.17193389

Running the webserver in Docker container

1. Download image data and sql database from zenodo into a folder (for instance /home/your_username/). Make sure the database is at the same level as the folder of the individual antibiotics i.e.

/home/
└──your_username/
    ├── database.db
    ├── Ampicillin/
    ├── Gentamicin/
    ├── Chloramphenicol/
    ├── Tetracycline/
    └── Ciprofloxacin/

2. Download docker image from github repo

   docker pull ghcr.io/baroudlab/adclab:latest

3. Activate docker on your computer (see https://docs.docker.com/engine/install/ for installation)

   docker run -p 8080:5000 -v /home/your_username/:/home/default-user/data/ ghcr.io/baroudlab/adclab:latest

5. Access the application at: http://localhost:8080

Brief explanation of the scripts not associated to figures: These are intermediate calculation and lead at the end lead the tables provided

1. CreateTrainingSetFromDatabase: Construct table with training dataset from the webserver and creates tif-crops used for training. (Requires running webserver.)
2. Training: Trains pytorsch's resnet34 on the training data. Creates network weights and tables for the performance metrics. (Requires Crops produced)
3. ApplyTraining: Applies trained neural network to all data and constructs tables with labels. (Requires running webserver and Crops)
4. ExtractClassProbs: Extracts single morphology probabilities from tables
5. Extracts coocurance morphology probabilities from tables

Tables provided

tables/ contains all the tables used in the figures