Machine Learning-Enhanced Terahertz-Colorimetric Dual-Mode Metadevice Based on Metallic Nanonetworks for microRNA Panel Assays

Abstract

Early diagnosis and accurate staging are essential for improving treatment efficacy and clinical outcomes of pancreatic cancer (PC), but remain challenging because of limited early PC-specific biomarkers and high-performance detection methods. In this paper, we developed a machine learning (ML)-enhanced terahertz (THz)–colorimetric dual-mode metadevice based on metallic nanonetworks for multiplexed microRNA (miRNA) detection and early PC diagnosis. Clinical sample measurements and bioinformatics analysis identified miRNA-21, miRNA-1290, and miRNA-320a as candidate PC biomarkers. Target miRNAs triggered the assembly of DNA-functionalized Fe and Au nanoparticles (NPs) with PEI-modified CuS NPs into ternary Fe/Au/Cu nanonetworks, generating THz and colorimetric signals through their dielectric and peroxidase-like properties. This target-triggered assembly enabled three-miRNA panel analysis without enzyme-mediated nucleic acid amplification and achieved attomolar-level miRNA profiling. Six ML algorithms were integrated to develop predictive models for PC diagnosis and staging. Linear discriminant analysis (LDA) achieved diagnostic accuracies of 97.8% (95% CI: 88.2–99.9%) in the training cohort (n = 45) and 96.7% (95% CI: 82.8–99.9%) in the validation cohort (n = 30), with 100% sensitivity for early PC in both cohorts. Compared with CA19-9, CA125, and CEA, the multiplexed miRNA-based dual-mode metadevice showed higher efficacy (area under the curve, AUC: 1.00 vs 0.905), supporting its potential for clinical miRNA-based liquid biopsy.