Machine learning assisted wearable antifouling sensor for reliable sweat analysis under dynamic conditions

Abstract

Wearable sweat sensors face persistent reliability challenges from biofouling and dynamic physiological variations during real-world use. To overcome these dual limitations, we introduce a paradigm-shifting co-design strategy integrating an antifouling catalytic hydrogel with machine learning-driven dynamic compensation. Our platform features: (1) A peptide composite hydrogel incorporating Au-PdNPs/rGO nanohybrids and engineered hydrophilic peptides, achieving exceptional antifouling performance (8.3 % signal loss in undiluted sweat) while providing specific catalytic activity toward uric acid (UA); (2) An artificial neural network (ANN) trained on 1217 experimental datasets spanning simultaneously varying physiological conditions. This synergistic approach enables three critical advances: First, the hydrogel's extreme hydrophilicity (9.01° contact angle) prevents surface fouling by sweat, maintaining electrode activity. Second, the ANN dynamically decouples environmental interference from target signals, overcoming the “single-variable optimization” limitation of conventional sensors. Third, the real sweat validation demonstrates accurate UA prediction, significantly outperforming static calibration methods and matching ELISA accuracy. By unifying antifouling material engineering with adaptive machine learning, this work establishes a new framework for reliable wearable biosensing, achieving prediction accuracy (R² = 0.9989) under real-world dynamic conditions.