Robust Logistic Regression-based Diagnosis Method of Prostate Cancer Using Optimized Feature Selection on Race...

Area of Science:

• Genomics
• Computational Biology
• Oncology

Background:

• Prostate cancer (PCa) incidence disproportionately affects Black men.
• Current diagnostic methods like prostate-specific antigen (PSA) testing lack specificity.
• Machine learning models often overlook racial variations in gene expression, impacting diagnostic fairness.

Purpose of the Study:

• To develop a race-aware PCa detection framework.
• To improve diagnostic accuracy and fairness using optimized feature selection.
• To identify race-specific biomarkers for enhanced PCa diagnosis.

Main Methods:

• Analysis of RNAseq-Count-STAR and clinical data from TCGA (554 patients).
• Feature selection pipeline integrating Differential Gene Expression, ROC analysis, and Gene-Set Enrichment Analysis.
• Development of a 9-gene logistic regression model, trained on White data and validated on Black data.

Main Results:

• The 9-gene model achieved 95% accuracy in White and 96.8% accuracy in Black populations.
• Fairness analysis showed minimal disparity between racial groups (4% difference in demographic parity, p=0.518).
• Biologically informed feature selection enhanced model accuracy and interpretability.

Conclusions:

• A race-specific PCa detection framework utilizing targeted biomarkers improves diagnostic accuracy.
• The framework addresses misclassification risks inherent in race-agnostic models.
• Emphasizes the importance of race-aware gene expression in machine learning diagnostics for precision medicine in PCa care.