Predicting protein cascade expression from H&E images.

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Biomarker discovery

Background:

• Protein expression in oncogenic pathways is crucial for cancer development.
• Predicting downstream protein signals is essential for understanding cancer progression.
• Current AI models often predict single proteins, lacking insight into signal propagation.

Purpose of the Study:

• To develop and evaluate an AI model (CellViT) for predicting downstream protein expression from whole-slide images (WSIs).
• To compare the performance of cellular-level ViT (CellViT) against patch-level Vision Transformers (ViT) for protein expression prediction.
• To investigate the utility of apoptosis and DNA damage/repair (DDR) cascades for predicting protein expression.

Main Methods:

• Utilized Reverse Phase Protein Array (RPPA) and WSI data from the TCGA-BRCA dataset.
• Developed a cellular-level Vision Transformer (CellViT) for predicting five key apoptosis cascade proteins.
• Compared CellViT performance against patch-level ViT models on a regression task.

Main Results:

• Patch-level ViT models achieved R-squared values < 0.1, indicating poor predictive performance.
• CellViT achieved R-squared values > 0.1 across all five test folds.
• The apoptosis cascade demonstrated significantly higher predictive performance than the DDR cascade.

Conclusions:

• CellViT demonstrates superior performance in predicting downstream protein expression compared to patch-level ViT models.
• Morphologically indicative biological cascades, like apoptosis, are more effective predictors of protein expression.
• This approach enhances the understanding of oncogenic signaling pathways through digital pathology.