Machine Learning-Driven Multiplexed Biomarker Detection with Polymer-Enhanced Electrochemical Sensors

Abstract

Biomarkers in sweat and saliva offer a promising avenue for non-invasive health monitoring. Electrochemical sensors have the potential to measure such biomarkers simultaneously. However, they are limited in discriminating individual biomarkers in mixtures, as redox potentials often overlap, resulting in current signatures that cannot be deconvoluted. This study focuses on differentiating biomarkers using orthogonal sensing materials combined with machine learning. We introduce a flexible electrochemical sensor array comprising carbon flower electrodes modified with poly(vinylidene fluoride) (PVDF) or poly(4-vinylpyridine) (P4VP) for the detection of estradiol (E2), ascorbic acid (AA), serotonin (5-HT), and melatonin (Mel). The two polymers act by altering the redox potential and current response of each biomarker, thereby enhancing signal diversity and enabling peak separation. Using multi-output regression models on 450 single and mixture measurements, the array accurately predicts concentrations (R2 = 0.95) over a wide dynamic range spanning nanomolar to micromolar levels. Polymer-resolved analysis reveals that PVDF-modifications enhance E2 and Mel detection, while P4VP-modifications improve AA and 5-HT quantification, highlighting the benefit of complementary orthogonal sensing electrodes. This finding is further supported by feature attribution analysis, which shows that the machine learning model relies on polymer-specific electrochemical signatures, directly linking improved performance to distinct polymer-analyte interactions. Overall, these results demonstrate that combining polymer-modified orthogonal electrodes with machine learning enables accurate, multiplexed sensing in complex mixtures, advancing selective detection strategies for future sensor platforms.