Stardust: improving spatial transcriptomics data analysis through space-aware modularity optimization-based...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) integrates tissue imaging with RNA sequencing for spatially resolved gene expression.
• Clustering is crucial for ST analysis, grouping data points (spots) based on similarity, impacting downstream analyses.
• Current ST clustering methods consider transcriptional similarity and sometimes spatial information, but the benefit of combining them is unclear.

Purpose of the Study:

• To introduce Stardust, a novel clustering method for spatial transcriptomics data.
• To evaluate the effectiveness of combining spatial and transcriptomic information in clustering.
• To compare Stardust's performance against state-of-the-art methods using stability and biological coherence metrics.

Main Methods:

• Developed Stardust, a method allowing manual or automatic tuning of spatial and transcriptomic data integration.
• Implemented a parameter-free version of Stardust where spatial influence is dynamic.
• Validated Stardust on 10x Genomics ST datasets, assessing cluster stability and biological coherence.

Main Results:

• Stardust effectively combines spatial and transcriptomic information for clustering.
• The method demonstrates improved cluster stability, meaning spots remain consistently clustered under perturbations.
• Stardust achieves greater biological coherence in identified clusters compared to existing methods.

Conclusions:

• Stardust offers an easy-to-use methodology for spatial transcriptomics clustering.
• The approach allows flexible control over the influence of spatial information.
• Stardust consistently yields more stable clustering results across different tissues than current state-of-the-art methods.