AI-enabled virtual spatial proteomics from histopathology for interpretable biomarker discovery in lung cancer

Area of Science:

• Biomedical imaging
• Computational pathology
• Proteomics

Background:

• Spatial proteomics offers high-resolution protein expression mapping but faces clinical translation challenges like cost and scalability.
• Standard histopathology slides lack detailed protein expression data crucial for understanding disease.
• AI integration into pathology is needed to bridge the gap between imaging and molecular insights.

Purpose of the Study:

• To develop and validate an AI model (HEX) for computationally generating spatial proteomics profiles from Hematoxylin and Eosin (H&E) stained histopathology slides.
• To assess the performance of HEX in predicting protein expression and enhancing prognostic and predictive accuracy in cancer.
• To explore the utility of integrating AI-derived virtual spatial proteomics with H&E images for improved clinical decision-making.

Main Methods:

• Developed HEX, an AI model trained on 819,000 histopathology image tiles and matched protein expression data from 382 tumor samples.
• Validated HEX's accuracy in predicting 40 diverse biomarkers, comparing its performance against alternative methods.
• Implemented a multimodal data integration approach combining H&E images with AI-derived virtual spatial proteomics.

Main Results:

• HEX accurately predicted the expression of 40 biomarkers, demonstrating superior performance over existing methods for protein prediction from H&E images.
• Multimodal integration using HEX improved prognostic accuracy by 22% and immunotherapy response prediction by 24-39% in six independent non-small-cell lung cancer cohorts (2,298 patients).
• Identified specific tumor-immune niches, such as T helper and cytotoxic T cell co-localization in responders and immunosuppressive cell aggregates in non-responders, as predictive of therapeutic outcomes.

Conclusions:

• HEX offers a low-cost, scalable method for analyzing spatial biology from standard histopathology slides.
• AI-derived virtual spatial proteomics integrated with H&E imaging significantly enhances prediction of patient outcomes and treatment response.
• HEX facilitates the discovery and clinical translation of interpretable biomarkers, paving the way for precision medicine.