SpaICL: image-guided curriculum strategy-based graph contrastive learning for spatial transcriptomics clustering.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics enables gene expression analysis within tissue context.
• Understanding tissue organization requires integrating multi-modal data.
• Existing methods face challenges in accurately delineating spatial domains.

Purpose of the Study:

• To introduce SpaICL, an image-guided graph contrastive learning framework for spatial transcriptomics clustering.
• To enhance the delineation of spatial functional domains by integrating diverse data types.
• To improve the accuracy and robustness of spatial transcriptomics analysis.

Main Methods:

• SpaICL utilizes an image-guided curriculum strategy and graph contrastive learning.
• It integrates gene expression, spatial coordinates, and histological image features.
• A dual cross-attention mechanism and curriculum learning module are employed to refine embeddings and mitigate over-smoothing.

Main Results:

• SpaICL achieved superior clustering performance on five benchmark spatial transcriptomics datasets.
• The framework effectively delineated spatial functional domains.
• Outperformed existing baseline methods in spatial transcriptomics clustering.

Conclusions:

• SpaICL provides a powerful new tool for spatial transcriptomics clustering.
• The framework demonstrates significant potential for downstream analytical applications.
• The integration of multi-modal data and advanced learning strategies enhances biological insights.