MUSE: A Multi-slice Joint Analysis Method for Spatial Transcriptomics Experiments

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) enables large-scale multi-slice data generation, increasing statistical power.
• Cross-slice inconsistencies and data quality variations pose significant analytical challenges in ST data.

Purpose of the Study:

• To develop a computational framework, MUSE, for multislice joint embedding, spatial domain identification, and gene expression imputation.
• To address limitations in current ST analysis, particularly cross-slice inconsistencies and data variability.

Main Methods:

• MUSE utilizes a two-module architecture for cross-slice alignment and data harmonization.
• Optimal transport is employed for cell alignment across slices, preserving spatial continuity.
• An alignment loss refines integration, enabling lower-quality data to benefit from higher-quality slices.

Main Results:

• MUSE demonstrated superior performance in cross-slice consistency, spatial domain identification, and gene expression imputation.
• The framework consistently outperformed existing methods across 12 real and 48 simulated ST datasets.
• MUSE generates virtual neighbors to enrich contextual information and mitigate data sparsity.

Conclusions:

• MUSE provides a robust and extensible framework for integrating multiple ST slices, advancing spatial gene expression analysis.
• The open-source software package promotes accessibility and adoption for complex biological systems research.
• MUSE enhances the applicability of single-slice methods to multi-slice ST data analysis.