Advancing liquid biopsy with qPCR validation: A machine learning-based model for lung cancer detection.

e20025 Background: Lung cancer remains a leading cause of cancer-related mortality worldwide. Early detection significantly improves survival rates. In our previous study, we developed a machine learning-driven diagnostic model using miRNA profiles from 148 plasma samples (74 cases, 74 controls) via next-generation sequencing (NGS), achieving a 10-fold cross-validation accuracy of 97.7%, sensitivity of 98.5%, and specificity of 97.0%. However, validation data was limited, and the high cost of NGS posed challenges for widespread application. This study addresses these gaps by incorporating isolated validation with 29 new cases and 30 controls, followed by qPCR verification of miRNA biomarkers to evaluate cost-effective alternatives for clinical use. Methods: We recruited 29 lung cancer cases (stage 0-II: 26; III-IV: 3) and 30 healthy controls from Chang Gung Memorial Hospital. Plasma-derived miRNA was sequenced using the QIAseq miRNA Library kit on the Illumina NextSeq550 platform. RNA-seq data were analyzed via the QIAGEN RNA-seq Analysis Portal 5.0. For qPCR validation, Ct-values of six miRNA biomarkers and one endogenous control were obtained using Magnetic Induction Cycler (MIC) real-time PCR system (Biomolecular Systems, Australia). Normalized values were subjected to differential analysis, and mean ± 2 standard deviations from controls were used as thresholds to determine outliers. Results: The NGS-based model validated on isolated data achieved an accuracy of 91.2% (95% CI: 81.1%-96.2%), sensitivity of 85.2% (95% CI: 67.5%-94.1%), and specificity of 96.7% (95% CI: 83.3%-99.4%). qPCR analysis confirmed the significance (p < 0.05) of 4/6 miRNA biomarkers. Outlier analysis revealed that 96.7% of controls had ≤1 outlier miRNA, whereas 79.3% of cancer cases exhibited ≥2 outliers. The proportion of cancer samples with ≥2 outliers was significantly higher than controls, underscoring the potential of these miRNAs in cancer detection. Conclusions: Our findings validate the robustness of the NGS model and demonstrate the feasibility of transferring NGS-identified biomarkers to qPCR, a cost-effective and scalable diagnostic method. qPCR's affordability and performance make it a promising tool for widespread lung cancer screening, potentially benefiting a broader population.