Connecting Histopathology Imaging and Proteomics in Kidney Cancer through Machine Learning.

Area of Science:

• Oncology
• Bioinformatics
• Computational Pathology

Background:

• Proteomics data offer molecular insights into cancer biology and diagnosis.
• Histopathology imaging is a standard clinical tool for cancer diagnosis.
• The link between large-scale proteomics and histopathology images is not well understood.

Purpose of the Study:

• To investigate the predictive relationship between proteomics and histopathology imaging in clear cell renal cell carcinoma.
• To apply machine learning, including deep neural networks, to integrate these data types.
• To identify molecular features in proteomics that correlate with imaging-based predictions.

Main Methods:

• Utilized proteomics and histology imaging datasets from clear cell renal cell carcinoma patients (CPTAC).
• Employed machine learning models, specifically deep neural networks, for data analysis.
• Correlated diagnostic protein sets with predictions from an imaging classification model.

Main Results:

• Established robust correlations between specific diagnostic proteins and imaging-based predictions.
• Identified proteins significantly associated with histology predictions involved in immune response, extracellular matrix reorganization, and metabolism.
• Demonstrated that genes encoding these proteins also recapitulate biological associations with imaging predictions via gene-protein correlations.

Conclusions:

• Machine learning can effectively integrate proteomics and histology imaging data in cancer research.
• Proteomic signatures correlate with histopathology-based predictions, offering insights into cancer mechanisms.
• This integrative approach provides a foundation for new diagnostic and research opportunities in oncology.