A machine learning-optimized bivalent aptasensor based on exonuclease Ⅰ amplification for trace okadaic acid detection

Consumption of okadaic acid (OA)-contaminated seafood endangers human health, highlighting the need for sensitive quantitative detection. Herein, a convenient and highly sensitive bivalent aptasensor was developed for fluorescence detection of OA in aquatic products. The aptasensor was constructed using a bivalent aptamer (B-Apt) as the recognition element, DNA-silver nanoclusters (cDNA‑B‑Apt‑cDNA‑AgNCs) as the fluorescent probe, and Exonuclease Ⅰ (Exo Ⅰ)-assisted target recycling for signal amplification. To improve accuracy and reduce matrix interference, a machine learning (ML) approach was integrated for fluorescence spectral analysis. Principal component analysis (PCA) was applied for dimensionality reduction, followed by linear regression for concentration prediction, and the K-nearest neighbors (KNN) algorithm for classification. Under optimal conditions, the aptasensor achieved a limit of detection of 16.18 pM and a linear range of 500 pM-150 nM, with high specificity against other marine toxins and superior sensitivity over previously reported methods. Spike recovery tests in seafood samples (clam, prawn, squid, kelp, yellow catfish) yielded recoveries of 93.09–106.59 % and relative standard deviations below 9.06 %. The PCA-based model demonstrated excellent concentration prediction performance (R 2 = 0.9906), and the KNN model achieved 100% classification accuracy for OA at different concentrations. This bivalent aptasensor combined with interpretable ML offers a reliable and practical platform for hazardous OA monitoring in food safety applications.