X-SPATIO: An Explanatory Deep Learning Pipeline for the Prediction and Visualization of Spatially Resolved Biomarker...

Area of Science:

• Computational pathology
• Molecular pathology
• Precision oncology

Background:

• Histopathology is crucial for cancer diagnosis but lacks routine molecular insights.
• Spatial omics platforms are costly and limited in scalability.
• Current computational models obscure subregion-specific morpho-molecular relationships, especially in heterogeneous triple-negative breast cancer (TNBC).

Purpose of the Study:

• To develop a computational pipeline, X-SPATIO, that links routine H&E morphology with spatial mRNA and protein expression.
• To enable cost-effective inference of spatial biomarker expression from histopathology images.
• To establish a foundation for discovery and precision oncology in TNBC.

Main Methods:

• Trained X-SPATIO on H&E regions of interest paired with spatially-resolved transcriptomic and proteomic data from GeoMx Digital Spatial Profiler.
• Utilized a multiple-instance learning approach to capture morpho-molecular associations.
• Generated spatio-morphologic attention maps to identify predictive tissue regions.

Main Results:

• X-SPATIO demonstrated strong performance for spatial biomarkers (AUC 0.79–0.97).
• Generated attention maps aligned with known biological spatial patterns and tissue organization.
• Successfully integrated spatial molecular data with routine histopathology.

Conclusions:

• X-SPATIO enables cost-effective inference of spatial biomarker expression from H&E images.
• The pipeline facilitates biologically grounded discovery in precision oncology for TNBC.
• X-SPATIO bridges the gap between histopathology and spatial molecular data.