Machine-Learning-Assisted Single-Entity Nanoimpact Electrochemistry with Enzyme-Responsive Deagglomeration for Sensitive Trypsin Detection

Abstract

Single-entity nanoimpact electrochemistry (NIE) has emerged as a powerful technique for biological target detection due to its ultrahigh sensitivity, rapid response, and single-event resolution. However, achieving reliable quantification of nonelectroactive biomolecules (e.g., trypsin) remains challenging because of difficulties in signal interpretation and the propensity for nonspecific aggregation in complex biological matrices, which often lead to false-positive results. Herein, we report a trypsin-responsive hydroxyl sulfhydryl polyethylene glycol peptide (HS-PEG-peptide) conjugate that specifically induces the deagglomeration of silver nanoparticle (AgNP) aggregates to achieve highly sensitive and selective detection of trypsin combined with machine-learning-assisted NIE signal analysis. This sensing strategy is based on the reversible aggregation and deagglomeration of AgNPs regulated by the RRK peptide and the HS-PEG-peptide conjugate. In the presence of trypsin, the HS-PEG-peptide conjugate is specifically cleaved, releasing HS-PEG and restoring its deagglomeration capability, which results in a concentration-dependent increase in collision frequency, as confirmed by chronoamperometric recordings. To overcome the limitations of manual data processing, we developed a Python-based machine-learning algorithm for automated collision peak detection, feature extraction, and collision frequency calculation. This algorithm not only eliminates subjective bias in data interpretation but also enables rapid analysis of large-scale collision data sets. Using random forest regression and principal component analysis, we further demonstrate that the extracted collision features can effectively distinguish different degrees of AgNP aggregation induced by various RRK concentrations. This method exhibits a linear detection range for trypsin from 0.25 to 300 ng/mL with a detection limit of 0.25 ng/mL and shows excellent selectivity over other interfering enzymes. By synergistically integrating smart responsive materials with machine-learning analysis in single-entity electrochemistry, this work provides a robust and automated platform for biomarker detection with high reliability and minimal human intervention.