3d-OT: a deep geometry-aware framework for heterogeneous slices alignment of spatial multi-omics.

Area of Science:

• Computational biology
• Spatial multi-omics
• Bioinformatics

Background:

• Spatial multi-omics technologies are advancing rapidly, offering insights into tissue heterogeneity.
• Current computational methods face challenges in integrating diverse molecular and spatial data.
• A need exists for advanced tools to analyze complex spatial biological information.

Purpose of the Study:

• To develop a deep geometry-aware framework (3D-OT) for integrating spatial geometric and multi-omics data.
• To improve feature extraction, spatial domain identification, and alignment of heterogeneous spatial slices.
• To enable a comprehensive understanding of spatial multi-omics data and biological tissue complexity.

Main Methods:

• 3D-OT framework utilizing modality fusion representation for spatial slice alignment.
• Soft correspondence optimal transport to handle nonrigid deformations in slice alignment.
• Chamfer distance for performance quantification.

Main Results:

• 3D-OT outperforms existing methods in capturing anatomical details in mouse brain cortex.
• Successfully tracks nonrigid deformations in heart and neural crest tissues across resolutions.
• Constructs a 3D spatiotemporal trajectory of mouse embryonic development.

Conclusions:

• 3D-OT provides a powerful computational tool for deciphering spatial complexity in biological tissues.
• The framework enables a comprehensive understanding of existing spatial multi-omics data.
• Advances the field of spatial biology by improving data integration and analysis.