AMICI: Attention Mechanism Interpretation of Cell-cell Interactions

Cross-attention-based cell-cell interaction inference from spatial transcriptomics data.

Contents:

• AMICI Quick Start Tutorial
  • Load the and split the data
  • Setup and train AMICI
  • AMICI’s downstream interpretation
• API Reference
  • Main Classes
  • Interpretation Modules
  • Callbacks
  • Utilities
• Changelog
  • Unreleased
  • Added
  • Version 0.1

Installation

You need to have Python 3.10 or newer installed on your system. If you don’t have Python installed, we recommend installing Mambaforge.

Install the latest release of amici-st from PyPI:

pip install amici-st

Or install the latest development version:

pip install git+https://github.com/azizilab/amici.git@main

Quick Start

import anndata
from amici import AMICI

# Load your spatial transcriptomics data
adata = anndata.read_h5ad("./adata.h5ad")

# Setup the data for AMICI
AMICI.setup_anndata(adata, labels_key="cell_type", coord_obsm_key="spatial")

# Create and train the model
model = AMICI(adata, **model_params)
model.train()

# Get attention patterns
attention_patterns = model.get_attention_patterns()
attention_patterns.plot_attention_summary()

Citation

If you find our work useful, please cite our preprint:

@article{Hong2025.09.22.677860,
    title = {AMICI: Attention Mechanism Interpretation of Cell-cell Interactions},
    author = {Hong, Justin and Desai, Khushi and Nguyen, Tu Duyen and Nazaret, Achille and Levy, Nathan and Ergen, Can and Plitas, George and Azizi, Elham},
    doi = {10.1101/2025.09.22.677860},
    journal = {bioRxiv},
    publisher = {Cold Spring Harbor Laboratory},
    year = {2025},
}