Machine Learning-Assisted Multiplex Profiling of Plasma-Derived Exosomal miRNAs for Alzheimer's Disease Classification

Abstract Accurate diagnosis of Alzheimer's disease (AD) remains challenging due to its multifactorial nature and the limitations of current diagnostic methods. Exosome-derived microRNAs (miRNAs) have emerged as promising minimally invasive biomarkers, owing to their stability in peripheral biofluids and their ability to reflect molecular alterations associated with neurodegeneration. In this work, we report an ORCA–Cas assay that integrates a one-pot ligation/rolling circle amplification step with a subsequent CRISPR/Cas12a detection step for the sensitive and specific detection of AD-associated exosomal miRNAs in plasma samples. Four candidate miRNAs (miR-16-5p, miR-23a-5p, miR-574-5p, and miR-361-5p) were evaluated in plasma-derived exosomes from AD patients and healthy controls using both RT–qPCR and ORCA–Cas platforms. In an independent cohort (n = 27 per group), individual miRNAs exhibited modest discriminative performance, highlighting the need for multivariate integration. A random forest model was subsequently employed to integrate signals from the four-miRNA panel, achieving an area under the curve (AUC) of 0.877 with RT–qPCR data and an AUC of 0.923 with ORCA–Cas-derived data. These results demonstrate that combining amplification-assisted CRISPR-based miRNA detection with multivariate modeling improves the diagnostic utility of plasma-derived exosomal miRNA profiles. The ORCA–Cas assay provides a simple, multiplex-capable, and sensitive platform that supports a multi-marker profiling strategy for plasma-based AD molecular profiling and classification.