Engineering the Interfacial Charge Transfer Dynamics by Plasmonic S-Scheme Heterojunctions for Machine-Learning-Assisted Dual-Mode Immunoassays

The development of photoelectrochemical (PEC)-coupled dual-mode biosensors, combined with multivariate regression analysis, is pivotal for advancing point-of-care disease marker diagnostics. Herein, we present a machine learning (ML)-powered dual-channel immunoassay. This platform integrates plasmonic TiO₂@NH₂-MIL-125/Au S-scheme photoelectrode with nanoconfined fluorescent CdSe@ZIF-8 probes. The optimized photoelectrode exhibits a remarkable photocurrent density of 10.29 μA/cm², representing a 581% enhancement over that of pristine TiO₂ (1.77 μA/cm²). Systematic investigation of interfacial charge transfer dynamics via density functional theory and in situ electron paramagnetic resonance analysis reveals synergistic plasmonic near-field coupling and robust built-in electric fields within the TiO₂@NH₂-MIL-125/Au. Leveraging this advanced photoelectrode, a smartphone-compatible PEC-coupled dual-mode biosensor is self-constructed and achieves exceptional detection capabilities for cardiac troponin I (cTnI), with an ultralow limit of detection of 6.01 fg/mL. Dual-mode signals (photocurrent and fluorescence RGB values) are processed by designing a recursive correlation framework incorporating a random forest algorithm for feature optimization. A convolutional neural network trained on multivariate data sets from 224 samples generates a robust regression model for cTnI quantification. This model demonstrates outstanding predictive ability, characterized by high accuracy (<i>R</i>² = 0.9966) and low prediction errors (5%). This study establishes an intelligent, field-deployable platform that merges dual-mode sensing with ML analytics for transformative point-of-care diagnostics.