Haruka Resolves Perturbation Response Heterogeneity in Spatial Cell Niches.

Area of Science:

• Computational biology
• Spatial omics
• Machine learning

Background:

• Tissue remodeling is complex, involving both conserved architecture and condition-specific changes.
• Dissecting these components requires advanced computational tools for spatial analysis.

Purpose of the Study:

• To present Haruka, a novel spatially aware contrastive learning framework.
• To enable the identification of salient (condition-specific) and background (shared) spatial domains across diverse tissue samples and experimental conditions.

Main Methods:

• Haruka integrates contrastive variational inference with microenvironment reconstruction.
• It learns spatial-context-informed embeddings capturing perturbation effects and local neighborhood context.

Main Results:

• Haruka outperforms state-of-the-art methods in detecting spatially heterogeneous responses on simulated and real datasets.
• It successfully distinguished immunotherapy responders in melanoma, traced fibrosis progression in lung tissue, and mapped resistant microenvironments in lung cancer.

Conclusions:

• Haruka offers a generalizable framework for spatial contrastive analysis.
• It facilitates systematic dissection of tissue organization, cellular plasticity, and microenvironmental remodeling.