Mosaic integration of spatial multi-omics with SpaMosaic

Area of Science:

• Spatial biology
• Multi-omics data integration
• Computational biology

Background:

• Spatial multi-omics technologies generate large, heterogeneous datasets.
• Integrating these datasets is crucial for building comprehensive biological atlases.
• Existing methods struggle with batch effects and modality gaps.

Purpose of the Study:

• To develop SpaMosaic, a novel computational tool for integrating diverse spatial multi-omics data.
• To enable accurate spatial domain identification and imputation of missing modalities.
• To create a modality-agnostic and batch-corrected latent space for unified analysis.

Main Methods:

• SpaMosaic utilizes contrastive learning and graph neural networks.
• It builds a modality-agnostic, batch-corrected latent space.
• Benchmarking involved simulated and real-world datasets across various tissues and omics types.

Main Results:

• SpaMosaic outperformed existing methods in identifying coherent spatial domains.
• It effectively reduced noise and mitigated batch effects.
• The tool accurately imputed missing modalities, revealing novel regulatory links in mouse brain data.
• SpaMosaic demonstrated scalability for large-scale integration.

Conclusions:

• SpaMosaic offers a versatile framework for unifying heterogeneous spatial omics data.
• It advances the construction of comprehensive multimodal spatial atlases.
• The tool enables deeper biological insights by inferring cross-modality relationships.