Automated Mycobacterial Growth Detection Algorithm (AMyGDA)

This is a python3 module that takes a photograph of a 96 well plate and assesses each well for the presence of bacterial growth (here Mycobacterial tuberculosis). Since each well contains a different concentration of a different antibiotic, the minimum inhibitory concentration, as used in clinical microbiology, can be determined.

A paper describing the software and demonstrating its reproducibility and accuracy is available from Microbiology.

The development of this software was funded by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre (BRC) to aid the CRyPTIC project.

Philip W Fowler

philip.fowler@ndm.ox.ac.uk

27 January 2020

Citing

Please cite

Automated detection of bacterial growth on 96-well plates for high-throughput drug susceptibility testing of Mycobacterium tuberculosis
Philip W Fowler, Ana Luiza Gibertoni Cruz, Sarah J Hoosdally, Lisa Jarrett, Emanuele Borroni, Matteo Chiacchiaretta, Priti Rathod, Sarah Lehmann, Nikolay Molodtsov, Clara Grazian, Timothy M Walker, Esther Robinson, Harald Hoffmann, Timothy EA Peto, Daniela Maria M. Cirillo, E Grace Smith, Derrick W Crook
Microbiology (2018) 164:1522-1530 doi:10.1099/mic.0.000733

Installation

This is python3; python2 will not work. Installation is straightforward using the included setup.py script. First clone the repository (or download it directly from this GitHub page)

$ git clone https://github.com/philipwfowler/amygda.git

This will download the repository, creating a folder on your computer called amygda/. If you only wish to install the package in your $HOME directory (or don't have sudo access) issue the --user flag

$ cd amygda/
$ python setup.py install --user

Alternatively, to install system-wide

$ sudo python setup.py install

The setup.py will automatically looks for the required following python packages and, if they are not present, will install them, or if they are an old version, will update them.

The information below is only included in case this process does not work. The prerequisites are

• numpy and scipy. Your python installation often includes numpy and scipy. To check, issue the following in a terminal

    $ python -c "import numpy"
    $ python -c "import scipy"
  

  If you see an error, indicating numpy and/or scipy is not installed, please install the scipy stack by following these instructions. -matplotlib. If your python installation includes numpy and scipy, there is a good chance it also includes matplotlib. Again to check

    $ python -c "import matplotlib"
  

You can find installation instructions here.

• opencv-python. This can be installed using standard python tools, such as pip

    $ pip install opencv-python
  

  AMyGDA was developed and tested using version 3.4.0 of OpenCV. If you do not have sudo access on your machine you can install this (and any other python module) in your $HOME directory using the following command

    $ pip install opencv-python --user
  

• datreant. This provides a neat way of storing and discovering metadata for each image using the native filesystem. It is not essential for the operation of AMyGDA, but the code would need re-factoring to remove this dependency. Again it can be installed using pip

    $ pip install datreant 
  

  Note that datreant works best if each image is containing within its own folder. datreant automatically stores all metadata associated with each image within two JSON files in a hidden .datreant folder in the same location as the input file.

Tutorial

The code is structured as a python module; all files for which can be found in the amygda/ subfolder.

$ ls
LICENCE.md                   amygda/                       setup.py
README.md                    examples/

(You may see other folders like build/ if you are run the setup.py script. To run the tutorial move into the examples/ sub-folder.

$ cd examples/
$ ls
analyse-plate-with-amygda.py plate-configuration/          sample-images/

analyse-plate-with-amygda.py is a simple python file showing how the module can be used to analyse a single image. The fifteen images shown in Figure S1 in the Supplement of the accompanying paper (see above) are provided so you can reconstruct Figures S2, S3, S4 & S12. The images are organised as follows

$ ls sample-images/
01 02 03 04 05 06 07 08 09 10 11 12 13 14 15

$ ls sample-images/01/
image-01-raw.png

To process and analyse a single image using the default settings is simply

$ analyse-plate-with-amygda.py --image sample-images/01/image-01-raw.png

And should take no more than 10 seconds. No output is written to the terminal, instead you will find a series of new files have been written in the samples-images/01 folder.

$ ls -a sample-images/01/
.datreant/
image-01-arrays.npz
image-01-filtered.png
image-01-mics.txt
image-01-processed.png
image-01-raw.png

• The hidden .datreant/ folder contains two JSON files. categories.json contains all the MICs and other metadata about the plate and both can be automatically discovered and read using the datreant module to make systematic analyses simpler.
• image-01-mics.txt contains the same information as the JSON file but in a simpler format that is easier for humans to read.
• image-01-arrays.npz contains a series of numpy arrays that specify e.g. the percentage growth in each well
• image-01-raw.png is the original image of the plate.
• image-01-msf.jpg is a JPEG of the plate following mean shift filtering
• image-01-clahe.jpg is a JPEG of the plate following mean shift filtering and then a Contrast Limited Adaptive Histogram Equalization filter to improve contrast and equalise the illumination across the plate.
• image-01-final.jpg is a JPEG of the plate following both the above filtering operations and a histogram stretch to ensure uniform brightness.
• image-01-growth.png adds some annotation; specifically the locations of the wells are drawn, each well is labelled with the name and concentration of drug and wells which AMyGDA has classified as containing bacterial growth are highlighted with a coloured circle.

To see the other options available for the analyse-plate-with-amygda.py python script

$ analyse-plate-with-amygda.py --help
usage: analyse-plate-with-amygda.py [-h] [--image IMAGE]
                                [--growth_pixel_threshold GROWTH_PIXEL_THRESHOLD]
                                [--growth_percentage GROWTH_PERCENTAGE]
                                [--measured_region MEASURED_REGION]
                                [--sensitivity SENSITIVITY]
                                [--file_ending FILE_ENDING]

optional arguments:
  -h, --help            show this help message and exit
  --image IMAGE         the path to the image
  --growth_pixel_threshold GROWTH_PIXEL_THRESHOLD
			the pixel threshold, below which a pixel is considered
			to be growth (0-255, default=130)
  --growth_percentage GROWTH_PERCENTAGE
			if the central measured region in a well has more than
			this percentage of pixels labelled as growing, then
			the well is classified as growth (default=2).
  --measured_region MEASURED_REGION
			the radius of the central measured circle, as a
			decimal proportion of the whole well (default=0.5).
  --sensitivity SENSITIVITY
			if the average growth in the control wells is more
			than (sensitivity x growth_percentage), then consider
			growth down to this sensitivity (default=4)
  --file_ending FILE_ENDING
			the ending of the input file that is stripped. Default
			is '-raw'

To analyse all plates, you can either use a simple bash loop

	$ for i in 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15; do
		analyse-plate-with-amygda.py --image sample-images/$i/image-$i-raw.png
	  done;

Alternatively if you have GNU parallel installed you can use all the cores on your machine to speed up the process.

	$ find sample-images/ -name '*raw.png' | parallel --bar analyse-plate-with-amygda.py --image {}

To delete all the output files, thereby returning sample-images/ to its clean state, a bash script is provided. Use with caution!

$ cd samples-images/

$ ls 01/
image-01-mics.txt
image-01-arrays.npz                                        image-01-clahe.jpg
image-01-filtered.jpg                                      image-01-raw.png
image-01-growth.jpg					   image-01-msf.jpg

$ bash remove-output-images.sh

$ ls 01/
image-01-raw.png

Licence

The software is available subject to the terms of the attached academic-use licence.

Adapting for different plate designs

AMyGDA is written to be agnostic to the particular design of plate, or even the number of wells on each plate. The concentration (or dilution) of drug in each well is defined by a series of plaintext files in

config/

For example the drugs on the UKMYC5 plate is defined within the

config/UKMYC5-drug-matrix.txt

file and looks like.

BDQ,KAN,KAN,KAN,KAN,KAN,ETH,ETH,ETH,ETH,ETH,ETH
BDQ,AMI,EMB,INH,LEV,MXF,DLM,LZD,CFZ,RIF,RFB,PAS
BDQ,AMI,EMB,INH,LEV,MXF,DLM,LZD,CFZ,RIF,RFB,PAS
BDQ,AMI,EMB,INH,LEV,MXF,DLM,LZD,CFZ,RIF,RFB,PAS
BDQ,AMI,EMB,INH,LEV,MXF,DLM,LZD,CFZ,RIF,RFB,PAS
BDQ,AMI,EMB,INH,LEV,MXF,DLM,LZD,CFZ,RIF,RFB,PAS
BDQ,AMI,EMB,INH,LEV,MXF,DLM,LZD,CFZ,RIF,RFB,PAS
BDQ,EMB,EMB,INH,LEV,MXF,DLM,LZD,CFZ,RIF,POS,POS

Adding a new plate design is simply a matter of creating new files specifying the drug, concentration and dilution of each well. Note that changing the number of wells at present also involves specifying the well_dimensions when creating a PlateMeasurement object. Currently this defaults to (8,12) i.e. a 96-well plate in landscape orientation. As an example, the configuration files for the UKMYC6 plate, which is the successor to the UKMYC5 plate, are included although all the provided examples are of UKMYC5 plates.