SANNO: A Graph-Transformer Enhanced Optimal Transport Tool for Spatial Transcriptomic Annotation.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics enables single-cell resolution analysis of tissue heterogeneity.
• Accurate cell type annotation is crucial but challenging for spatial transcriptomics data.
• Current methods often fail to leverage spatial information and identify novel cell types.

Purpose of the Study:

• Introduce SANNO, a novel approach for cell type annotation in spatial transcriptomics.
• Enable concurrent identification of both known and novel cell types.
• Improve accuracy and robustness of cell annotation by integrating spatial data.

Main Methods:

• Employs Optimal Transport (OT) for cell type identification.
• Utilizes a graph-Transformer module to model spatial coordinates and gene expression.
• Features a dual-strategy classifier with Unbalanced Optimal Transport (UOT) and a self-supervised OT module.
• Incorporates an entropy-based re-weighted loss function for enhanced prediction confidence.

Main Results:

• SANNO surpasses state-of-the-art methods in intra- and cross-spatial dataset annotation.
• Demonstrates superior performance in identifying novel cell types.
• Shows strong results in annotating cells from single-cell RNA sequencing (scRNA-seq) data.

Conclusions:

• SANNO offers a versatile and powerful tool for cell annotation in both spatial and single-cell transcriptomics.
• Effectively addresses limitations of existing methods by incorporating spatial information.
• Facilitates more comprehensive understanding of cellular composition and spatial organization in tissues.