Photoelectrochemical Detection of Watermelon Sucrose Metabolic Enzyme RNA Based on NiO/NiS/N-rGO Composite and CatBoost Calibration.

An improved CatBoost calibration model integrating adaptive gradient clipping, Huber weighted loss, and Bayesian hyperparameter optimization was developed to drive quantitative detection of watermelon sugar-promoting enzyme RNA via a multisource information fusion framework of "sensor response-environmental parameters-target concentration", which integrates Huber weighted loss to address data heteroscedasticity, adaptive gradient clipping to suppress gradient explosion and overfitting, and Bayesian hyperparameter optimization to ensure parameter optimality. This model was applied to an enzyme-free photoelectrochemical (PEC) sensing platform constructed from a cDNA@NiO/NiS/N-rGO block-like hierarchical composite. In-situ grown NiO/NiS heterojunctions on N-rGO form interlocked honeycomb-pore and nanoneedle architectures, creating a p-n junction array (NiO p-type/NiS intrinsic/N-rGO n-type) that simultaneously enhances conductivity and active-center density. The sensor exhibits a detection limit of 0.067 pM and a broad linear range from 1 pM to 10 nM. By effectively overcoming data heteroscedasticity and overfitting through the improved CatBoost algorithm, the calibration model achieves a root mean square error (RMSE) of 1.85 pM, a coefficient of determination (R2) of 0.9946, and a mean absolute percentage error (MAPE) of 6.4%, corresponding to an approximately 23% improvement in detection accuracy over the standard CatBoost. The sensor further demonstrates excellent selectivity, stability, and reproducibility, offering technical support for in-situ monitoring of watermelon ripeness and precise quality control of fruit products.