Integrating miRNA profiling and machine learning for improved prostate cancer diagnosis.

Area of Science:

• Biomarkers
• Oncology
• Molecular Diagnostics

Background:

• Prostate cancer (PCa) diagnosis is challenging due to symptom overlap with benign prostatic hyperplasia (BPH) and limitations of PSA tests.
• Existing diagnostic tools have high false-positive rates, necessitating improved methods for accurate PCa detection.

Purpose of the Study:

• To investigate the diagnostic potential of microRNA (miRNA) biomarkers in whole blood for differentiating PCa from BPH.
• To evaluate the efficacy of machine learning (ML) models in analyzing miRNA expression data for enhanced diagnostic accuracy.

Main Methods:

• Prospective cohort study analyzing whole blood miRNA expression using reverse transcription polymerase chain reaction.
• Development and validation of a random forest ML model utilizing miRNA expression levels and ratios.
• Bioinformatics analysis to identify cancer-related pathways associated with selected miRNAs.

Main Results:

• Specific miRNAs (miR-21-5p, miR-141-3p, miR-221-3p) were identified as significant discriminators between PCa and BPH.
• The ML model achieved 77.42% accuracy and 0.78 AUC in verification, and 74.07% accuracy and 0.75 AUC in validation.
• miRNA expression ratios demonstrated superior sensitivity and specificity compared to PSA testing.

Conclusions:

• Whole blood miRNA profiling integrated with ML offers a promising non-invasive approach for improving PCa diagnostic precision.
• The identified miRNAs are biologically relevant, associated with key cancer pathways like PD-L1/PD-1 and androgen receptor signaling.
• Further research with larger populations and standardized protocols is required for clinical translation of miRNA-based diagnostics.