STransfer: a transfer learning-enhanced graph convolutional network for clustering spatial transcriptomics data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics analysis is crucial for understanding tissue architecture.
• Existing methods often overlook inter-slice similarities in multi-slice datasets.
• Accurate spatial structure capture is fundamental for biological insights.

Purpose of the Study:

• To develop a novel transfer learning framework for spatial transcriptomics.
• To address limitations in current methods by modeling inter-slice similarities.
• To enhance clustering accuracy and reduce manual annotation in spatial transcriptomics.

Main Methods:

• Proposed STransfer framework combining graph convolutional networks (GCNs) and positive pointwise mutual information (PPMI).
• Utilized an attention-based module for fusing multi-graph features into unified node representations.
• Developed low-dimensional embeddings encoding gene expression and spatial context.

Main Results:

• STransfer effectively models local and global spatial dependencies.
• The framework successfully transfers knowledge from labeled to unlabeled tissue slices.
• Achieved superior clustering accuracy and spatial modeling compared to state-of-the-art methods.

Conclusions:

• STransfer offers a robust solution for spatial transcriptomics data analysis.
• The method enhances the understanding of spatial gene expression patterns.
• STransfer improves efficiency by reducing manual annotation efforts.