krentzd/ai4ab

Deep learning recognises antibiotic mode of action from brightfield images

This repository contains model weights for the paper "Deep learning recognises antibiotic modes of action from brightfield images". See the GitHub repository for usage instructions.

Usage

Clone this repository with git clone https://github.com/krentzd/ai4ab.git and navigate to the model folder. Models trained on different sets of plates, imaging channels or bacterial species can then be loaded as follows:

from models import AvgPoolCNN

model = AvgPoolCNN.from_pretrained(
            species='Ecoli',                       # Select bacterial species
            channels=['BF', 'Hoechst'],            # Select imaging channels
            replicate=2,                           # Model replicate (different training/testing plates)
        )

The repository includes the following pretrained models with four replicates for each setting:

	E. coli ( Ecoli )	K. pneumoniae ( Kpneumoniae )
Brightfield ( BF )	✓	✓
FM4-64FX ( FM4 )	✓
Hoechst 3334 ( Hoechst )	✓
Brightfield ( BF ) FM4-64FX ( FM4 )	✓
Brightfield ( BF ) Hoechst 3334 ( Hoechst )	✓
FM4-64FX ( FM4 ) Hoechst 3334 ( Hoechst )	✓
Brightfield ( BF ) FM4-64FX (FM4 ) Hoechst 3334 ( Hoechst )	✓

The repository also contains models from four additional experiments:

• Trained on 16-bit images: experiment='16bit'
• Trained on MoA labels: experiment='MoA'
• Trained without resizing the input image: experiment='no-resize'
• Trained without data augmentation: experiment='no-data-aug'

To, for example, access a model trained on MoA labels, specify the experiment argument as follows:

model = AvgPoolCNN.from_pretrained(
            species='Ecoli',                  
            channels=['BF'],                  
            replicate=2,
            experiment='MoA'                 
        )