m2ST: dual multi-scale graph clustering for spatially resolved transcriptomics.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Spatial clustering is crucial for analyzing spatial transcriptomics data.
• Current graph neural network methods struggle with single-scale analysis, limiting feature discriminative power.
• Tailored clustering methods are needed to improve spatial domain annotation accuracy.

Purpose of the Study:

• To propose m2ST, a novel dual multi-scale graph clustering method for spatial transcriptomics.
• To address limitations of single-scale analysis and improve spatial domain annotation.
• To enhance the discriminative power of extracted features for clustering.

Main Methods:

• m2ST employs a multi-scale masked graph autoencoder for multi-scale representation extraction.
• A random masking mechanism and scaled cosine error are used for knowledge distillation.
• A tailored multi-scale clustering framework integrates scale-common and scale-specific information, with Shannon entropy for dynamic scale adjustment.

Main Results:

• m2ST extracts multi-scale representations from spatial transcriptomic data.
• The method achieves robust annotation performance by integrating scale-specific and scale-common information.
• Extensive experiments show m2ST outperforms existing methods on multiple datasets.

Conclusions:

• m2ST offers a superior approach to spatial domain clustering in transcriptomics.
• The dual multi-scale strategy effectively captures biological insights across different resolutions.
• This method advances the analysis of complex biological mechanisms through spatial transcriptomics.