The high worldwide mortality and disproportionate impact of cardiovascular diseases have emerged as the most significant global health burden, unfortunately, unmet by the traditional detection methods. Therefore, developing a rapid, sensitive, selective, and rugged biosensor for the precise classification/quantification of cardiac biomarkers is a stepping stone for the future generation of cardiac healthcare. We demonstrate a facile, time-efficient, and scalable biosensor for classifying the FDI approved gold standard cardiac biomarker Troponin-I (cTnI) in untreated human serum matrix, built-on 2-D SnS2 and 1-D MWCNT composite transducer and decision-tree based explainable machine learning (ML) algorithm. The proposed methodology is further enhanced using an inimitable Operating-Voltage-Selection-Algorithm (OVSA), which boosts ML accuracy to ∼100%. The near-perfect classification is realized by strategically incorporating this two-step algorithm-first the OVSA, then the heuristic and ML approaches on the selected dataset. Dynamic concentrations of the biomarker (100 fg/mL to 100 ng/mL) are estimated with high sensitivity, ∼71 (ΔR/R) (ng/mL)-1cm-2 and low limit of detection (0.02 fg/mL), aiding to the prediction and prognosis of acute myocardial infarction. The hyperparameter tuning and feature engineering improve the decision process of the ML algorithm, fostering robustness against data variability. Feature importance indices, namely the Gini index, Permutation Importance, and SHAP values, portray 'Voltage' as the most important feature, further justifying our insight into the OVSA. The biosensor's specificity, selectivity, reproducibility and stability are effectively demonstrated with the sampling to result reporting time of just 20 min, establishing it as a potential candidate for clinical testing.
Keywords: Biomarker; Chemiresistive biosensor; Feature importance; ML Explainability; Machine learning; SnS(2)-CNT Composite.
Copyright © 2022 Elsevier B.V. All rights reserved.