Deep Transfer Learning Links Benign Glands to Prostate Cancer Progression via Transcriptomics

Area of Science:

• Oncology
• Molecular Biology
• Bioinformatics

Background:

• The field effect describes cellular changes predisposing to cancer, but is underexplored in high-resolution omics.
• Prostate cancer progression involves complex interactions within the tumor microenvironment.

Purpose of the Study:

• To investigate the field effect in prostate cancer using deep learning and spatial transcriptomics.
• To identify molecular alterations in morphologically benign tissues associated with cancer aggressiveness.

Main Methods:

• Utilized the Diagnostic Evidence Gauge of Single Cells (DEGAS) deep transfer learning framework.
• Analyzed prostate cancer spatial transcriptomics and integrated single-cell transcriptomics with deep learning image analysis.
• Quantified MSMB protein (PSP-94) expression using immunohistochemistry on patient tissues.

Main Results:

• DEGAS identified benign glands with reduced MSMB expression and upregulated genes linked to antigen presentation and aggressive neoplasms.
• Altered immune-cell infiltration was observed in the tumor microenvironment.
• Lower MSMB protein expression in morphologically normal tissues correlated with increased risk of distant metastasis.

Conclusions:

• Subtle molecular and immune changes in ostensibly normal prostate tissue, representing a field effect, contribute to aggressive disease progression.
• These findings highlight the importance of analyzing the broader tissue context in understanding cancer aggressiveness.