AMICI: Attention Mechanism Interpretation of Cell-cell Interactions

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics offers insights into cellular interactions within tissues.
• Current analysis tools lack dynamic and interpretable methods for estimating cell communication and spatial ranges.
• Understanding cell-cell communication is crucial for deciphering tissue organization and function.

Purpose of the Study:

• To develop an interpretable attention framework, AMICI, for analyzing spatial transcriptomic data.
• To enable dynamic and interpretable estimation of cell-cell interactions and their spatial scales.
• To link intercellular communication patterns to downstream gene programs.

Main Methods:

• Developed AMICI, an interpretable attention framework.
• Jointly estimated interaction length scales and resolved sender-receiver subpopulations.
• Linked communication patterns to gene expression programs.

Main Results:

• AMICI successfully recovered ground-truth interactions in semi-synthetic datasets.
• Identified gene programs associated with cell communication in the mouse cortex.
• Revealed length-scale-dependent tumor-immune signaling influencing estrogen receptor (ER) programs in breast cancer.

Conclusions:

• AMICI provides a powerful and interpretable method for analyzing cell-cell communication in spatial transcriptomics.
• The framework enhances understanding of tissue-level biological processes, including development and disease.
• AMICI's ability to link communication to gene programs opens new avenues for biological discovery.