ResSAT: Enhancing Spatial Transcriptomics Prediction from H&E- Stained Histology Images with Interactive Spot...

Area of Science:

• Computational biology
• Genomics
• Medical imaging analysis

Background:

• Spatial transcriptomics (ST) provides high-resolution RNA quantification.
• Hematoxylin and eosin (H&E) images are crucial for medical diagnosis and reveal tissue structures.
• Existing methods for predicting spatial gene expression from H&E images often neglect spatial relationships.

Purpose of the Study:

• To develop a framework that generates spatially resolved gene expression profiles from H&E images.
• To improve the prediction of spatial gene expression by incorporating tissue structure and spatial relationships.
• To offer a cost-effective and rapid alternative for ST profiling.

Main Methods:

• Introduction of ResSAT (Residual networks - Self-Attention Transformer) framework.
• Utilizing residual networks to capture tissue structures from H&E images.
• Employing a self-attention transformer to enhance the prediction of spatially resolved gene expression.

Main Results:

• ResSAT significantly outperforms existing methods in predicting spatial gene expression.
• The framework effectively captures tissue structures and enhances prediction accuracy.
• Benchmarking conducted on 10× Visium datasets demonstrated ResSAT's superior performance.

Conclusions:

• ResSAT offers a promising approach for generating spatially resolved gene expression profiles from H&E images.
• The method has the potential to reduce the costs associated with ST profiling.
• ResSAT enables the rapid acquisition of numerous gene expression profiles, accelerating research.