In brief

QuestionCan the sensor materials be designed to make machine learning better at separating overlapping biomarkers?
ResultAcross 450 single- and mixed-analyte measurements, multi-output regression achieved an overall R² of about 0.95 for four biomarkers.
Why it mattersThe study co-designs electrochemical response diversity and the prediction model instead of asking AI to rescue an unchanged sensor.
Machine Learning-Driven Multiplexed Biomarker Detection with Polymer-Enhanced Electrochemical Sensors
Adrian L. M. Duesselberg et al. · ACS Electrochemistry · Published 12 May 2026
DOI: 10.1021/acselectrochem.6c00079

The electrochemical problem

Multiplexed electrochemical sensing becomes difficult when biomarkers oxidize at similar potentials. If several analytes produce overlapping peaks on the same electrode, there may not be enough independent information for reliable concentration estimation.

This paper takes a co-design approach: change the electrochemical signatures with different polymer interfaces, then let machine learning combine the resulting channels.

What the researchers did

The sensor array uses carbon-flower electrodes modified with either PVDF or P4VP. The two polymers shift redox potentials and alter current responses differently for estradiol, ascorbic acid, serotonin, and melatonin.

The authors collected 450 measurements from single-analyte and mixed-analyte samples and trained multi-output regression models to predict all four concentrations simultaneously.

What they found

The reported overall R² is about 0.95 across the four-analyte problem, over concentrations spanning nanomolar to micromolar ranges. Polymer-resolved analysis showed complementary behavior: PVDF improved the signatures for some analytes while P4VP improved others.

Feature-attribution analysis further indicated that the machine-learning model was using polymer-specific electrochemical signatures rather than only a generic current magnitude.

Why this matters

The important design choice is to create orthogonal information before modeling. Machine learning cannot reliably separate analytes if the measurement contains no distinguishing information; the polymer interfaces deliberately make the signals more different.

This is a useful blueprint for Echem + AI sensing: engineer the electrode so the data become more informative, then use ML to decode the richer response.

One important caution

The reported performance comes from defined single- and multi-analyte mixtures. Real sweat, saliva, or other biofluids introduce fouling, matrix effects, variable ionic composition, and additional redox-active species.

The result is therefore best read as a strong multiplexed-sensing architecture that still requires validation for long-term and real-matrix use.

Primary source

Adrian L. M. Duesselberg et al. Machine Learning-Driven Multiplexed Biomarker Detection with Polymer-Enhanced Electrochemical Sensors. ACS Electrochemistry (2026). DOI: 10.1021/acselectrochem.6c00079.

Research-note facts and quantitative values were checked against the publisher record linked above. The “Why this matters” and “One important caution” sections are EchemAI editorial interpretation.