Interpretable AI model reveals spatial codes of cell recruitment in tissues.

SPACER, an interpretable multi-instance learning model, improves recovery of gene-level cell recruitment signals by 4–21-fold across 37 spatially resolved transcriptomics datasets spanning human tumors and mouse heart.

• Why it matters: Spatial transcriptomics can map where cells are located, but it often cannot explain which molecular signals recruit them to specific tissue sites.
• What they did: SPACER decodes gene-level recruitment rules from SRT data using a fully interpretable multi-instance learning framework, then was benchmarked across 37 SRT datasets.
• The result: In tumor and mouse heart analyses, SPACER links tumour-expressed mucins to physical barriers to T cell engagement, associates local antigen presentation with T cell engagement across cancer types, and identifies Cd4 + T cells as dominant effectors in severe immune checkpoint inhibitor-induced myocarditis.

AI summary of the abstract · may contain errors · not medical advice · read the paper.