DNAzyme-protease cascade amplified triple-modal biosensing platform with machine learning integration for specific...

Sugarcane pokkah boeng disease poses a serious threat to global sugar production, highlighting an urgent need for on-site detection tools that are both highly sensitive and operable without sophisticated instrumentation. Herein, we report a self-powered biosensing platform integrating electrochemical, colorimetric, and photothermal tri-modal readouts assisted by machine learning for ultrasensitive and specific detection of the pokkah boeng pathogen. The sensing interface is constructed using a MWCNT@ZIF-8/AuNPs nanocomposite, which facilitates electron transfer and offers abundant active sites. By leveraging a cascaded amplification strategy involving DNAzyme cleavage and Exonuclease III (ExoIII)-assisted recycling, the system achieves significant signal enhancement without requiring multiple enzyme systems or complex sequence designs. The incorporation of a G-quadruplex/hemin DNAzyme enables not only colorimetric signaling through TMB oxidation but also robust near-infrared photothermal conversion, thereby complementing the electrochemical and colorimetric modes. Furthermore, machine learning algorithms, including Linear Regression, Ridge Regression, Lasso Regression, and SGD Regressor, are employed to model the multi-modal data, substantially improving prediction accuracy and robustness through cross-validation. The proposed biosensor demonstrates exceptional sensitivity with detection limits of 20.41 aM (electrochemical), 25.36 aM (colorimetric), and 18.91 fM (photothermal), across broad linear ranges. Practical applicability is confirmed through spike-recovery assays in real sugarcane leaf extracts, yielding satisfactory recoveries (99.0-108.0%) and high reproducibility (RSD <5%). This work provides a reliable and portable strategy for early plant disease monitoring and showcases the promising integration of multimodal sensing with machine learning for agricultural diagnostics.