sisal 0.1.0

Beta-variational autoencoder for IMS (and other) data.

sisal

The SiSAL package is an approach based on the beta-variational autoencoder and kernel density estimation to dissect data along independent, uncertainty-aware, and interpretable (yet non-linear) latent axes. It includes a novel comparative-latent-traversal algorithm to translate latent findings back into the original measurement context. You can find demonstrations in the /experiments folder. It includes an imaging mass spectrometry-based molecular imaging of human kidney and a synthetic dataset. The approach’s disentangling properties are shown to impose a latent space structure that separates signal strength from relative signal content, offering exceptional chemical insight.

Installing

You can install sisal from PyPI:

pip install sisal

or directly from source:

git clone https://github.com/vandeplaslab/sisal.git

pip install "."

Examples

If you are planning on running the examples, you can install the optional dependencies with:

pip install sisal[demo]

This will include additional packages such as jupyterlab and pooch to download data from Zenodo.

You can find the example data on Zenodo.

Contributing

Contributions are always welcome. Please feel free to submit PRs with new features, bug fixes, or documentation improvements.

git clone https://github.com/vandeplaslab/sisal.git

pip install -e .[dev]

Figures

Kidney imaging mass spectrometry datasets

Figure 1: Kernel density estimation of the latent space generated by the SiSAL approach together with the associated masks.

Figure 2 (latent traversal): Traversal example between the two highlighted regions.

Figure 3 (feature differentiators): The traversal algorithm extracts the most important molecular features between the two regions.