An Explanatory Deep Learning Pipeline for the Prediction and Visualization of Spatially Resolved Biomarker Expression...

Area of Science:

• Computational pathology
• Molecular imaging
• Cancer research

Background:

• Histopathology is crucial for cancer diagnosis but lacks routine molecular insights.
• Spatial omics platforms offer region-specific data but face cost and scalability limits.
• Current computational models often miss detailed spatial morpho-molecular connections, especially in heterogeneous cancers like TNBC.

Purpose of the Study:

• To develop X-SPATIO, a computational pipeline linking H&E morphology with spatial mRNA and protein expression.
• To enable cost-effective inference of spatial biomarker expression in routine histopathology.
• To support discovery and precision oncology in triple-negative breast cancer (TNBC).

Main Methods:

• Developed X-SPATIO, a spatially compatible computational pipeline.
• Trained the model on H&E regions and spatially-resolved omics data (GeoMx DSP).
• Utilized a multiple-instance learning approach to capture morpho-molecular associations and generate attention maps.

Main Results:

• X-SPATIO demonstrated strong performance for spatial biomarkers (AUC 0.79-0.97).
• Generated spatio-morphologic attention maps highlighting predictive tissue regions.
• Attention maps aligned with known biological patterns and tissue organization.

Conclusions:

• X-SPATIO effectively integrates spatial molecular data with histopathology.
• Enables cost-effective inference of spatial biomarker expression, crucial for TNBC.
• Provides a foundation for biologically grounded discovery and precision oncology in TNBC.