GALA: a unified landmark-free framework for coarse-to-fine spatial alignment across resolutions and modalities in...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Spatial transcriptomics alignment faces challenges from technical variations like geometric distortions and platform differences.
• Diverse alignment scenarios arise, including mismatched resolutions, cross-modality integration, and partial tissue coverage.

Purpose of the Study:

• To introduce GALA (Genetic Algorithm-guided Large Deformation Alignment), a novel framework to address spatial transcriptomics alignment challenges.
• To enable accurate and efficient multimodal data integration across different resolutions and modalities.

Main Methods:

• GALA employs a unified, landmark-free framework combining global affine transformation and local diffeomorphic deformation in a single optimization.
• Modality-aware rasterization harmonizes transcriptomic and histological data onto a shared grid.

Main Results:

• GALA successfully aligns spatial transcriptomic and histological data across resolutions and modalities in a landmark-free manner.
• Evaluations on human and mouse datasets demonstrate GALA's superior accuracy, computational efficiency, and biological interpretability compared to existing methods.
• The framework effectively handles both complete and partial tissue alignment scenarios.

Conclusions:

• GALA provides a robust and versatile solution for spatial transcriptomics alignment, overcoming limitations of previous methods.
• The framework facilitates more accurate and interpretable integration of multimodal spatial omics data.