flowCat is a classifier and visualizer for flow cytometry data. The classifier uses self-organized maps for representing raw ungated flow cytometry data. A CNN is then used for generating predictions.
The processing of a single sample involves:
- Transformation of FCS into SOM
- CNN prediction
- Generation of Saliency Mappings
- Potential embedding analysis and other batched methods
The SOM generation and classification both require tensorflow, so a recent NVIDIA GPU is preferable. The system has been internally used using a GTX1080 and a P40. If Video RAM is a limitation, reducing the batch size for the SOM generation might be beneficial.
Software dependencies needed are:
- tensorflow 1.12
Other python dependencies should be correctly installed by pip. Use that on the downloaded wheel file.
$ pip install /path/to/wheelInstallation in a conda environment is preferable for better management of dependencies.
Download miniconda and the following commands could be used to setup a conda environment.
$ conda activate
$ conda create -n flowcat python=3.6
$ conda activate flowcatAfterwards follow the main instructions for installing the flowCat package.
Clone this repository and then install via issuing:
$ python3 setup.py develop
Optionally this should be done in a specific python environment to avoid collisions with other versions of the same python libraries.
flowCat can be either used entirely as a command line utility or its components can be used in custom python scripts.
The command line utility can be used to manipulate properly formatted datasets and transform these into SOM datasets. Custom data sources and other formattings should be transformed using custom python scripts using flowCat as a library.
$ flowcat --help # overview of available commandsA main convenience library in flowcat.flowcat is available, containing a
simplified interface to the major functions required for classification and SOM
generation.
The main results from the paper can be generated using the
./flowCat_gen_results.sh script. If you have issues running the scripts please
contact the authors.
tensorflow SOM code adapted from cgorman.