Integrative cross-sample alignment and spatially differential gene analysis for spatial transcriptomics

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) technologies provide gene expression data with spatial context.
• Comparing multiple ST slices is difficult due to nonlinear distortions and limited overlap.
• Existing methods struggle with cross-platform and cross-sample comparisons.

Purpose of the Study:

• To introduce CODA, an integrative framework for cross-sample alignment and spatially differential gene analysis.
• To enable robust comparison and extraction of spatial gene expression patterns across diverse ST data.
• To overcome challenges in aligning and comparing ST slices from various platforms.

Main Methods:

• CODA learns a shared low-dimensional latent feature space across samples.
• It performs global affine alignment and transformer-based feature matching for domain identification.
• Local nonlinear refinements use large deformation diffeomorphic metric mapping for accurate alignment.

Main Results:

• CODA demonstrates strong performance in alignment accuracy, computational efficiency, and memory usage across ST platforms.
• The framework successfully identifies common spatial domains.
• CODA uncovers spatially informative genes related to normal and disease conditions via immunofluorescence and enrichment analysis.

Conclusions:

• CODA provides a robust and effective framework for spatial transcriptomics analysis.
• It significantly improves cross-sample alignment and comparison capabilities.
• The framework has broad applicability for uncovering spatially resolved biological insights.