Artificial Intelligence Design for Race-Based Prostate Cancer Stage Classification With Multilayer Perceptron:...

Area of Science:

• Genomic Medicine
• Computational Biology
• Oncology

Background:

• Prostate cancer progression varies significantly due to biological and racial factors.
• DNA methylation shows promise for early cancer detection, but its use in machine learning across diverse populations is limited.

Purpose of the Study:

• To develop a prostate cancer stage classifier using DNA methylation data and a multilayer perceptron (MLP) model for a predominantly White cohort.
• To evaluate the model's performance on other racial groups to highlight the need for race-specific models.

Main Methods:

• Processed TCGA-PRAD data using differentially methylated position (DMP) analysis to identify CpG sites correlated with cancer stages.
• Refined features using recursive feature elimination (RFE) and trained MLP models.
• Employed SHAP and LIME for model interpretation and identification of key DNA methylation features.

Main Results:

• The best model achieved 95% accuracy and 99% AUC on White training data using 90 features.
• Model performance significantly declined in racial minority groups due to sample imbalance and race-specific methylation patterns.
• Feature importance analysis revealed specific CpG sites driving model predictions.

Conclusions:

• A race-aware MLP model for prostate cancer staging using DNA methylation data was developed and optimized.
• SHAP and LIME confirmed the predictive relevance of selected CpG sites, enhancing model transparency.
• The study highlights the critical need for race-specific modeling strategies in cancer classification due to performance disparities across racial groups.