Kontext: Kernel-based method bridging spatial domain identification and cell-cell communication

The spatial arrangement of biological tissues consists of spatial domains - regions characterized by consistent genetic expression and physical traits. These domains are largely shaped by direct cell-to-cell communication. While pinpointing these spatial domains is a frequent requirement in spatial transcriptomics, many current strategies lack deeper understanding into the underlying mechanisms.

We introduce Kontext, a kernel-based method aimed at simultaneously modeling individual cells and their immediate surroundings. This facilitates the accurate discovery of spatial domains based on actual ligand-receptor interactions between cells. The method also yields supplementary data for exploring cellular microenvironments, spatial niches, and communication pathways from senders to receivers.

When tested against existing techniques, Kontext demonstrates superior performance in reproducing recognized domain annotations across multiple spatial transcriptomics platforms and biological samples, while simultaneously delivering superior computational efficiency and scalability. When applied to a Xenium temporal mouse heart cryoinjury dataset, Kontext provides a detailed account of the injury's spatial and temporal progression.

It also uncovers an previously unidentified continuous activity of the complement system in the interaction between fibroblasts and macrophages during the formation of scar tissue. Furthermore, in a massive 3D Open-ST dataset comprising a million cells from a human metastatic lymph node tissue, Kontext is used to integrate spatial domain identification with niche analysis.

This enables the definition of a boundary region at the interface between the tumor and its surrounding microenvironment. The study reveals unique communication patterns between this interface and the central tumor core.