A comprehensive comparison on clustering methods for multi-slice spatially resolved transcriptomics data analysis

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) enables gene expression and spatial pattern analysis within tissues.
• Clustering is crucial for ST data, revealing spatial organizations with shared characteristics.
• Multi-slice clustering methods are emerging for contiguous tissue sections.

Purpose of the Study:

• To comprehensively compare single-slice and multi-slice clustering methods for spatial transcriptomics data.
• To evaluate the impact of preprocessing techniques on clustering performance.
• To provide a practical guide for selecting appropriate clustering methods for multi-slice ST data.

Main Methods:

• Assessed seven single-slice and four multi-slice clustering algorithms.
• Utilized two simulated and four real spatial transcriptomics datasets.
• Investigated the effects of spatial coordinate alignment (e.g., PASTE) and batch effect removal (e.g., Harmony).

Main Results:

• Performance varied across clustering methods depending on dataset characteristics.
• Preprocessing techniques like spatial alignment and batch correction influenced clustering outcomes.
• Multi-slice methods showed potential for improved spatial domain detection in integrated analyses.

Conclusions:

• No single clustering method is universally optimal for all multi-slice ST data scenarios.
• Method selection should consider data complexity, biological questions, and preprocessing steps.
• This comparison serves as a valuable resource for researchers applying spatial transcriptomics.