abct

abct (abct.rubinovlab.net) is a MATLAB and Python toolbox for unsupervised learning, network science, and imaging/network neuroscience.

The toolbox includes three variants of global residualization, the Loyvain and co-Loyvain methods (for k-means, k-modularity, or spectral clustering of data or network inputs), as well as binary and weighted canonical and co-neighbor components, and m-umap embeddings. It also includes computation of degree centralities (first, second, and residual), dispersion centralities (squared coefficient of variation, k-participation coefficient), and network shrinkage.

See this reference for more details: Rubinov M. Unifying equivalences across unsupervised learning, network science, and imaging/network neuroscience. arXiv, 2025, doi:10.48550/arXiv.2508.10045.

Download

ABCT MATLAB ABCT Python GitHub

Interactive examples

The examples illustrate the main analyses in the above study, and can be run interactively as online Jupyter notebooks in Google Colab (You need a Google account). To run in Colab, click the Open in Colab button at the top of the page of each example.

Our example brain-imaging data come from the Human Connectome Project, a large brain-imaging resource.

NB: For illustrative ease, some example analyses are simplified versions of analyses in the original study.

Contact

Mika Rubinov: mika.rubinov at vanderbilt.edu