An Interpretable Multi-instance Learner Decodes Cellular Recruitment from Spatially Resolved Transcriptomics.

Area of Science:

• * Spatial biology and systems immunology.
• * Computational biology and bioinformatics.
• * Molecular mechanisms of cell trafficking and tissue microenvironments.

Background:

• * Studying cell recruitment and engagement in tissue microenvironments is crucial but challenging with traditional methods.
• * Genome-wide analysis of these processes requires advanced computational tools.
• * Spatially resolved transcriptomics (SRT) data offers a powerful resource for such investigations.

Purpose of the Study:

• * To develop a novel, interpretable deep learning model for analyzing SRT data.
• * To investigate the mechanisms of stromal and immune cell recruitment in tumors and myocarditis.
• * To uncover T cell localization features and differential responses in disease contexts.

Main Methods:

• * Development of "Spacer," a multi-instance deep learner for analyzing transcriptomics and spatial modalities of SRT data.
• * Application of Spacer to 17 high-definition and 20 low-definition SRT datasets.
• * Validation of Spacer's findings using immuno-peptidomics, spatial T cell receptor (TCR) sequencing, and single-cell sequencing.

Main Results:

• * Identified genes promoting T cell recruitment to tumors, distinct molecular programs for B cell and macrophage recruitment, and mucin-mediated T cell repulsion.
• * Uncovered T cell-intrinsic features governing localization, validated by spatial-TCR-seq.
• * Revealed CD4+ T cells are more responsive than CD8+ T cells in myocarditis, despite lower numbers.

Conclusions:

• * Spacer provides a new spatially resolved paradigm for studying cellular localization mechanisms *in situ*.
• * The study shifts from descriptive mapping to mechanistic discovery in spatial biology.
• * Findings offer novel insights into immune cell dynamics in tumors and inflammatory heart disease.