Adaptive Electrochemical Aptamer Sensing across Biofluid-Relevant pH Ranges via a Dual-Stage Machine Learning Framework.

Reliable electrochemical sensing can be hindered by environmental variability such as pH fluctuations, ionic strength, and nonspecific adsorption, which compromise signal fidelity and quantitative accuracy. Here, we present a dual-stage machine learning-assisted electrochemical framework that decouples environmental interference from concentration-dependent responses using ferrocene-labeled DNA-gold nanoparticle (Fc-DNA@AuNP) probes as a model aptamer platform. The redox-active nanostructure generated tunable differential-pulse voltammetry (DPV) signals, whose morphology and amplitude varied systematically across ionically standardized pH-calibration media covering biofluid-relevant acidic-to-weakly alkaline regimes. Machine-learning classifiers first identified pH conditions from shape-based electrochemical descriptors including peak potential, prominence, width, and signal-to-noise ratio, achieving over 85% accuracy across four pH-variable environments (pH 3.6-8.0). Subsequently, pH-specific regressors predicted carcinoembryonic-antigen (CEA) concentrations spanning five orders of magnitude, with quantitative performance evaluated using log-scale RMSE, MAE, and R2. The combined workflow enables adaptive interpretation of distorted electrochemical profiles without requiring additional internal standards or recalibration. This study provides an application-oriented pH-adaptive interpretation strategy for aptamer-based electrochemical sensors and establishes a transferable signal-decoding layer for subsequent complex-matrix and portable analytical validation.