Machine Learning-Integrated Raman Spectroscopy Identifies Race-Associated Biochemical Signatures in Prostate Cancer

Black men experience disproportionately higher prostate cancer incidence and mortality, yet the underlying biochemical contributors remain unclear. In this study, we integrate Raman spectroscopy with advanced multivariate and machine-learning methods to characterize molecular differences in clinical formalin-fixed, paraffin-embedded (FFPE) prostate tissues from Black and White patients. Raman spectra were corrected using ICA-PLS, wavelet-denoised, and unmixed with Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) to quantify cellular components. Random forest (RF) models were trained on denoised spectra to classify cancer versus control tissues. MCR-ALS revealed elevated protein, collagen, lipid, and nucleic acid signatures in tumors from Black patients, aligning with clinically observed aggressive disease phenotypes. RF classification achieved > 90% accuracy, 95% sensitivity, 85% specificity, and an AUC > 0.96, demonstrating robust diagnostic performance. These findings show that Raman spectroscopy integrated with computational analysis provides a powerful label-free approach to probe biochemical drivers of racial disparities in prostate cancer.