PromptSTG: prototype-guided prompting for few-shot spatial transcriptomics annotation

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell resolution spatial transcriptomics (scST) offers insights into tissue architecture and cell interactions.
• Accurate cell type annotation in scST is challenging due to scarce labels, high costs, and complex tissue heterogeneity, particularly in tumor microenvironments.

Purpose of the Study:

• To develop a robust few-shot learning framework for accurate cell type annotation in scST data.
• To address limitations in current annotation methods, especially in complex and heterogeneous tissues.

Main Methods:

• Proposed Prompt-guided Spatial Transcriptomics Graph (PromptSTG), a graph-based few-shot learning framework.
• Integrated spatial and transcriptomic features to model cellular neighborhoods and enable label propagation.
• Utilized a small set of labeled cells to annotate large unlabeled populations.

Main Results:

• PromptSTG demonstrated superior accuracy, robustness, and scalability in few-shot cell type annotation across multiple scST platforms and tissue types.
• The method successfully reconstructed spatially coherent tissue organization and identified rare cell populations.
• PromptSTG preserved global tissue structure and fine-grained cellular boundaries in complex environments, yielding consistent annotations.

Conclusions:

• PromptSTG provides a powerful tool for accurate and efficient cell type annotation in scST data, particularly in challenging few-shot scenarios.
• The framework's ability to handle complex tissues and identify rare cells enhances biological discovery from spatial transcriptomics.
• PromptSTG offers a scalable and robust solution for advancing tissue analysis and understanding cellular heterogeneity.