Innovative super-resolution in spatial transcriptomics: a transformer model exploiting histology images and spatial...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics maps tissue microenvironments but often lacks single-cell resolution due to platform limitations (e.g., Visium).
• Existing methods may require costly single-cell RNA sequencing (scRNA-seq) for high-resolution analysis.

Purpose of the Study:

• To introduce TransformerST, an unsupervised model for enhancing spatial transcriptomics resolution to the single-cell level.
• To enable cost-efficient, high-resolution spatial transcriptomics analysis by integrating histology images and gene expression data.

Main Methods:

• Developed TransformerST, an unsupervised model based on the Transformer architecture.
• Employed a vision transformer-based encoder for image-gene expression co-representation.
• Integrated spatial correlations using an adaptive graph Transformer module and a cross-scale graph network for super-resolution.

Main Results:

• TransformerST successfully elevates Visium data to single-cell granularity.
• The model demonstrates adaptability across various spatial transcriptomics platforms.
• Empirical evaluations confirm its accuracy in revealing tissue structures at the single-cell scale.

Conclusions:

• TransformerST is a pioneering tool for spatial transcriptomics, offering single-cell resolution.
• It optimally integrates gene expression and histology images, enhancing the understanding of tissue structure-function relationships.
• The unsupervised, reference-free approach makes high-resolution spatial transcriptomics more accessible.