Whole Slide Imaging-Based Prediction of TP53 Mutations Identifies an Aggressive Disease Phenotype in Prostate Cancer

Area of Science:

• Computational pathology
• Oncology
• Biomarker discovery

Background:

• Prostate cancer needs early prognostic biomarkers for metastatic potential.
• TP53 mutations are candidate biomarkers, but tumor heterogeneity complicates molecular profiling.
• Whole slide images (WSIs) offer potential for spatial profiling and mitigating heterogeneity.

Purpose of the Study:

• To assess WSIs as proxies for spatially resolved profiling.
• To evaluate WSIs as biomarkers for aggressive prostate cancer.
• To develop a deep learning model for predicting TP53 mutations from WSIs.

Main Methods:

• Developed TiDo, a deep learning model for predicting TP53 mutations from prostate cancer WSIs.
• Validated the model on an independent multifocal cohort at patient and lesion levels.
• Analyzed model predictions to understand false positives and associated phenotypes.

Main Results:

• TiDo achieved state-of-the-art performance in predicting TP53 mutations from WSIs.
• The model generalized successfully across different patient cohorts and lesions.
• False positive predictions were linked to TP53 deletions and downstream phenotypes associated with stromal composition.

Conclusions:

• WSIs can serve as proxies for molecular profiling and prognostic biomarkers in prostate cancer.
• Deep learning models on WSIs can identify aggressive disease phenotypes, even without perfect spatial mutation prediction.
• These models hold potential for elucidating tumor prognosis by capturing downstream phenotypes linked to aggressive disease biomarkers.