MICRON learns outcome-associated representations of spatial immune microenvironments.

Area of Science:

• Biomedical data analysis
• Computational biology
• Immunology

Background:

• Spatial imaging proteomics, like imaging mass cytometry, identifies immune microenvironments crucial for disease progression.
• Analyzing these complex datasets to link specific immune microenvironments to patient outcomes is challenging due to cell segmentation and sample heterogeneity.

Purpose of the Study:

• To introduce MICRON, a novel, segmentation-free, automated tool utilizing multiple-instance learning for identifying outcome-associated immune microenvironments.
• To enhance prognostic and diagnostic prediction accuracy using spatial imaging proteomics data.

Main Methods:

• Developed MICRON, a fully automated multiple-instance learning framework.
• Applied MICRON to spatial imaging proteomics data, focusing on brain cancer case studies.
• Generated importance maps to visualize key immune microenvironments and cell interactions.

Main Results:

• MICRON accurately identifies outcome-linked immune microenvironments without cell segmentation.
• The tool demonstrated improved prognostic and diagnostic prediction capabilities compared to existing methods.
• Case study in brain cancer revealed coordinated communication between astrocytes, NK cells, and macrophages associated with survival.

Conclusions:

• MICRON offers a robust, automated solution for analyzing spatial imaging proteomics data to uncover disease-driving immune microenvironments.
• The tool facilitates a deeper understanding of cell-cell communication in cancer, aiding prognostic and diagnostic efforts.
• MICRON is available as open-source software to support broader clinical and biological research.