python

ktest

This package implements kernel tests such as kernel Fisher Discriminant Analysis (kFDA) and Minimum Mean Discrepency (MMD) introduced in [1].

Note regarding code used in publication

The code for the analysis presented in [1] are available in a dedicated repository here.

The source code for experimental_data and squair scripts are based on the branch publication_genome_biology. The reversion scripts are based on the branch publication_genome_biology_reversion.

See https://github.com/AnthoOzier/ktest_experiment_genome_biology_2024 for more details.

Licensing and authorship

See the dedicated LICENSE.md and AUTHORS.md files.

Python package

Install

Latest development version:

pip install ktest@git+https://github.com/LMJL-Alea/ktest@main#subdirectory=python

Caution: For users of MacOS with x86 processors (before Apple switched to ARM processors), PyTorch (that is a dependency for ktest) stopped supporting this platform with version 2.3. As a consequence, you will be limited to Pytorch <2.3, and thus Python >=3.8,<3.13.

Tutorials

See the dedicated tutorials folder in the project root directory for notebook tutorials.

Usage

Example with a dataset used in [1]:

import pandas as pd
from ktest.tester import Ktest

# data
url = "https://raw.githubusercontent.com/LMJL-Alea/ktest/main/tutorials/v5_data/RTqPCR_reversion_logcentered.csv"
data = pd.read_csv(url, index_col=0)

meta = pd.Series(data = pd.Series(data.index).apply(lambda x : x.split(sep='.')[1]))
meta.index = data.index

# init ktest object
kt_1 = Ktest(
    data=data, metadata=meta, sample_names=['48HREV','48HDIFF'], nystrom=True
)
# run kfda test
kt_1.test()
print(kt_1)

# save object
kt_1.save("ktest_RTqPCR_reversion_logcentered.pkl", compress=True)

# load saved
kt_1 = Ktest.load("ktest_RTqPCR_reversion_logcentered.pkl", compressed=True)

References

[1] Anthony Ozier-Lafontaine, Camille Fourneaux, Ghislain Durif, Polina Arsenteva, Céline Vallot, Olivier Gandrillon, Sandrine Gonin-Giraud, Bertrand Michel, Franck Picard. Kernel-based testing for single-cell differential analysis. Genome Biol 25, 114 (2024). doi:10.1186/s13059-024-03255-1, hal-04214858