AFSC: A self-supervised augmentation-free spatial clustering method based on contrastive learning for identifying...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics enables gene expression analysis with spatial context.
• Spatial information is crucial for understanding cell communication, microenvironment interactions, and disease pathology.
• Identifying spatial domains through clustering is a key analysis step.

Purpose of the Study:

• To develop an improved spatial clustering method for spatial transcriptomics.
• To address limitations of existing contrastive learning methods that use data augmentation, potentially disrupting biological meaning.
• To propose a self-supervised method that effectively integrates spatial information and gene expression data.

Main Methods:

• Developed Augmentation-Free Spatial Clustering (AFSC), a self-supervised contrastive learning method.
• Constructed a contrastive learning module using Teacher and Student Encoders, avoiding negative pairs and data augmentation.
• Integrated an unsupervised clustering module trained jointly with the contrastive learning module.

Main Results:

• AFSC demonstrates strong performance in self-supervised spatial clustering across various spatial transcriptomics datasets and resolutions.
• The method effectively learns latent representations by integrating spatial information and gene expression.
• Learned representations are suitable for downstream tasks like visualization and trajectory inference.

Conclusions:

• AFSC offers a robust and biologically meaningful approach to spatial clustering in transcriptomics.
• The augmentation-free strategy preserves data integrity while leveraging spatial context.
• This method advances the analysis of spatial transcriptomics data for biological discovery.