Machine learning-enhanced multimodal electrochemical bioassay using multifunctional high-entropy alloy for complex mixtures

The detection of multiple trace analytes using single sensors is often impeded by the limited sensitivity of material and the interference form overlapping signals in complex mixtures. Here, we introduce an efficient and durable heterostructured high-entropy alloy (HEA) material, where non-noble HEA nanoparticles are used to disperse and stabilize Pt clusters (denoted as HEA@Pt). The HEA@Pt exhibits high sensitivity to three trace analytes (dopamine, uric acid, and paracetamol) during the electrochemical detection process, leveraging its multifunctional catalytic sensing capabilities for diverse mixtures. Additionally, to address the challenge of signal overlap, we integrate a recurrent neural network into multimodal sensing, combined with machine learning (ML) algorithms to accurately identify multiple analytes in mixtures. After five-fold cross-validation, the prediction accuracy deviations for dopamine, uric acid, and paracetamol were 3.20, 9.18 and 3.84, respectively, with goodness-of-fit values of 0.984, 0.992 and 0.990. The model achieved a prediction accuracy of 96.67% for unknown mixture samples, demonstrating robust generalization performance. This approach of multifunctional HEA combined with ML algorithms effectively overcomes detection errors caused by the complex detection of multiple chemical substances and the overlap of multiple response signals, thereby enabling accurate and reliable identification of multi-target analytes.