CellNiche represents cellular microenvironments in atlas-scale spatial omics data with contrastive learning.

Area of Science:

• Computational Biology
• Spatial Omics
• Single-cell Analysis

Background:

• Analyzing cellular microenvironments at scale is difficult due to coupled molecular identity, spatial context, and platform differences.
• Existing methods struggle to integrate diverse spatial omics data effectively.

Purpose of the Study:

• To develop a scalable framework, CellNiche, for identifying and characterizing cellular microenvironments from spatial omics data.
• To overcome challenges in integrating multi-platform spatial omics datasets and analyzing cellular neighborhoods.

Main Methods:

• CellNiche employs a contrastive learning approach using cell-centric spatial-proximity subgraphs.
• It combines spatial co-localization and molecular co-expression data to learn microenvironment-aware embeddings.
• The framework was validated on >10 million cells across multiple spatial omics platforms.

Main Results:

• CellNiche demonstrates improved data representation with increased training data and competitive performance in clustering and embedding quality.
• It efficiently integrates diverse spatial omics datasets, handling platform heterogeneity.
• The framework successfully identified conserved and sample-specific tumor and immune microenvironments in non-small-cell lung cancer (NSCLC) and integrated mouse brain atlases.

Conclusions:

• CellNiche provides a scalable and effective method for analyzing cellular microenvironments in spatial omics data.
• It enables robust integration of multi-platform datasets and accurate characterization of cellular neighborhoods.
• The framework has broad applications in understanding tissue architecture and disease mechanisms, as shown in cancer and brain atlases.