ZnO nanoflower-mediated paper-based electrochemical biosensor for perfect classification of cardiac biomarkers with physics-informed machine learning.

The widespread exposure of acute myocardial infarction globally demands an ultrasensitive, rapid, and cost-effective biosensor for troponin-I and T in a dynamic concentration range. Traditionally, the saturation of sensor response limits accurate prediction at high analyte concentrations, although this is seldom discussed in the literature. To address this research gap, we thematically report physics-informed analytical treatments with machine learning (PIML) on a paper-based electrochemical biosensor, taking advantage of low cost, flexibility, low sample volume, and ease of deployment. Owing to the well-known biosensing performances, ZnO nanoflowers, synthesized in-house with a hydrothermal procedure, are utilized for transduction purposes with three voltametric techniques: CV, DPV, and SWV. The exceptional surface coverage and high IEP of ZnO have contributed towards the realization of high sensitivity, and the monoclonal antibody-based bioreceptors ensured the enormous selectivity of the platform. Nevertheless, the traditional calibration approach for CV considers the peak current as the sensor parameter, which gets flattened at higher concentrations, thereby limiting reliability. Therefore, this issue is addressed by strategic analytical development by extracting the charge associated with a CV scan and employing this physics-informed feature in the machine learning (ML) model. Combining features generated from different electrochemical techniques in the ML model enhances data diversity by including comprehensive information. This unique approach towards data analysis led to achieving 100% accuracy and AUC scores for identifying cardiac troponin-I and T multiclass concentrations. We strongly believe that the proposed methodologies have a substantial potential for translation to any other related sensor applications.