miRNA Biomarkers in Prostate Cancer: Leveraging Machine Learning for Improved Diagnostic Accuracy

Abstract

Prostate cancer (PCa) diagnosis often relies on prostate-specific antigen (PSA) testing, but its high false-positive rates often lead to unnecessary biopsies. MicroRNAs (miRNAs) have emerged as promising non-invasive biomarkers for cancer detection due to their stability in biological fluid and disease specificity. Despite their potential, the clinical translation of miRNAs as non-invasive cancer biomarkers is hindered by several challenges - population-based variability, environmental Factors, methodological Inconsistencies, lack of standardization, normalization Issues, and complexity of the biological System. These factors significantly impact the consistency of miRNA expression readouts, particularly in terms of Ct-values, across different studies, which in turn affects the determination of cutoff values that are crucial in a diagnostic setup. This preliminary study offers a pilot demonstration for integrating miRNA biomarker expression with machine learning (ML), which can help identify patterns and improve classification, potentially reducing the reliance on fixed cutoff values in certain contexts and pave the path to wider clinical translation. We analyzed the expression of key miRNAs (miR-21-5p, miR-221-3p, and miR-141-3p) in blood samples from patients with PCa and benign prostatic hyperplasia (BPH). Utilizing a Random Forest classifier, we achieved an accuracy of 77.42%, a precision of 86.21%, a recall of 71.43%, and an AUC-ROC score of 0.78. The application of ML enabled us to leverage complex features, such as combinations and ratios of miRNA expression data, which enhanced the robustness and reliability of the diagnostic model. Additionally, bioinformatics analysis of the preferential features identified by the ML model confirmed the biological relevance of these miRNAs in PCa-related pathways, further supporting their potential as clinical biomarkers. In the future, ML is poised to significantly enhance diagnostic performance compared to traditional linear analyses of a limited set of biomarkers. While our study did not explore multiple populations or the effects of methodological variables, it highlights the potential of ML by demonstrating improved accuracy and eliminating the need for cutoff values. This capability could broaden the applicability of miRNA-based diagnostics, making them more reliable and actionable in clinical settings. However, to fully realize this potential, further validation with larger and more diverse cohorts is essential. Overall, this study lays the groundwork for utilizing ML-enhanced miRNA panels as powerful tools for the early and non-invasive diagnosis of PCa in future clinical practice. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=130 SRC="FIGDIR/small/618146v1_ufig1.gif" ALT="Figure 1"> View larger version (38K): org.highwire.dtl.DTLVardef@2eddd9org.highwire.dtl.DTLVardef@e6d6c8org.highwire.dtl.DTLVardef@11eea0forg.highwire.dtl.DTLVardef@989c15_HPS_FORMAT_FIGEXP M_FIG C_FIG