A workflow for training a logistic regression model that leverages both intrinsic transcriptomic signatures and spatial context—incorporating neighboring regions’ profiles and distance to the tissue edge (left box).
Its prediction pipeline is compatible with high-resolution spatial transcriptomics datasets (e.g., Visium HD, Xenium, MERFISH) via a pseudobulk strategy (right box).
Creat a conda environment and install
conda create --name tissuetypist_env python=3.10
conda activate tissuetypist_env
git clone https://github.com/Teichlab/TissueTypist.git
cd TissueTypist
pip install .
Please refer to the demo notebooks for examples using query datasets of various types and resolutions.
For imaging-based spatial transcriptomics data using targeted gene panels (e.g., Xenium, MERFISH), we recommend training a new model on the panel’s gene set. The demo notebooks include a dedicated training workflow with the built-in Visium reference dataset.
main_MERFISH.ipynb uses a publicly available dataset (download link provided in the notebook) and reports session information along with software dependencies. The expected runtime is approximately 10 minutes or less.