ProDiphy

Python package that implements probabilistic models to compare (sub-)populations.

Installation

Note: currently the package has not been deposited in a repository, so it needs to be installed through GitHub.

To install ProDiphy, conda is required (or miniconda).

First, clone the repository to create a local copy in the current directory.

git clone https://github.com/raeslab/prodiphy

Then navigate to the source code, create an environment using the yaml file in the folder docs/dev, activate it and install the local copy of the code.

cd prodiphy
conda create -n prodiphy python=3.12
conda activate prodiphy
pip install -e .

Note that creating the environment can take a few minutes as there are some fairly sizeable dependencies required for Prodiphy.

Quick Start

The following code snippet shows how to use the ProDir model to compare two populations. The model is used here to compare the prevalence of species in two ecosystems, one polluted and one unpolluted. The model will estimate the prevalence of each species in both ecosystems and provide a credible interval for the difference in prevalence.

from prodiphy import ProDir

if __name__ == "__main__":
    reference_counts = [100, 30, 20, 10]
    polluted_counts = [90, 20, 5, 10]
    
    labels = ["SpeciesA", "SpeciesB", "SpeciesC", "SpeciesD"]
    
    model = ProDir()
    model.fit(reference_counts, polluted_counts, labels)
    
    summary = model.get_stats()

For more details including the output of this example, see the ProDir documentation.

Usage

The Prodiphy package contains various models each with their own specific use case. The following models are currently available:

• ProDir: Model to compare the prevalence of specific classes in two populations. (Probabilistic alternative to a chi-square test)
• CorProDir: Model to compare the prevalence of specific classes in two populations while correcting for covariates.
• DeltaSlope: Model to compare the slope, intercept and spread of a linear regression between two groups.
• DMM: Dirichlet Multinomial Mixture model to detect clusters with different prevalence of classes.
• GMvM: Gaussian Multivariate Mixture model for clustering continuous multivariate data.
• ACORN: Adjusted CORrelations, Negative-binomial. Hierarchical model testing whether every feature in a count table is associated with a continuous marker, adjusted for covariates.

Contributing

Any contributions you make are greatly appreciated.

• Found a bug or have some suggestions? Open an issue.
• Pull requests are welcome! Though open an issue first to discuss which features/changes you wish to implement.

Contact

ProDiphy was developed by Sebastian Proost at the RaesLab. ProDiphy is available under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license.

For commercial access inquiries, please contact Jeroen Raes.