#=================================================================================================================================#
# READ ME: Automatic recognition of flow cytometric phytoplankton functional groups using Convolutional Neural Networks
#=================================================================================================================================#

This directory gathers the data, the model configurations and the results obtained in the paper "Automatic recognition of flow
cytometric phytoplankton functional groups using Convolutional Neural Networks" by Fuchs et al. (2022).

It is structured in the following way:

+---benchmark_data
|   +---SSLAMM
|   \---SWINGS
+---benchmark_models
|   +---SSLAMM
|   \---SWINGS
+---manual_heterogeneity
|   +---SSLAMM
|   \---SWINGS
\---three_month_data
    +---SSLAMM
    +---SWINGS
    \---reg_plot.py (the code to generate Figure 5 of the paper)

"manual_heterogeneity" contains the material related to the heterogeneity of the manual gating between experts for each data source (SSLAMM or SWINGS).
It contains the CytoClus4 selection sets and raw data files, the ARIs and CV coefficients, and the uncertainty maps presented in the paper (Figure 2).

"benchmark_data" contains the data used for models comparisons and the nomenclature used by the CNN.
The CNN has been trained on the Pulse shapes data and the other models on the Listmodes data.

"benchmark_models" gives the best specifications for each model.
The best hyperparameter values are given in the pickle files. The weights and the trained CNN are also provided. 

"three_month_data" contains the material related to the three months of CNN predictions on the SWINGS and SSLAMM data.
The raw Cytoclus4 .cyz files are given in "raw_data". 
The automatic and manually gated counts are given in the results folder. 

General guidelines on how to run the predictive pipeline are given in the phyto_curves_reco Github repository: 
https://github.com/RobeeF/phyto_curves_reco
The notebooks used to reproduce the results of the paper are also given in this repository.

If necessary, do not hesitate to contact Robin Fuchs for precisions or other intermediate outputs at robin.fuchs92@gmail.com.