Machine Learning with Neural Networks to Enhance Selectivity of Nonenzymatic Electrochemical Biosensors in Multianalyte Mixtures.

Nonenzymatic biosensors hold great potential in the field of analysis and detection due to long-term stability, high sensitivity, and low cost. However, the relative low selectivity, especially the overlapped oxidation peaks of biomarkers, in the biological matrix severely limits the practical application. In this work, we introduce an intelligent back-propagation neural network into nonenzymatic electrochemical biosensing to overcome the limitation of low selectivity for glucose and lactate detection. After simple electrodeposition and dropping modification, three working electrodes with distinct characters are fabricated and integrated into electrochemical microdroplet arrays for glucose and lactic acid detection. By analyzing chronoamperometry data from a standard mixture of glucose and lactate in varying concentrations, a database of highly selective detection can be simply established. The trained neural network model can reliably identify and accurately predict the concentration of glucose and lactic acid in the range of 0.25-20 mM with a correlation coefficient of 0.9997 in multianalyte mixtures. More importantly, the predicted results of serum samples are precise, and the relative standard deviation is less than 6.5%, proving the possible applicability of this method in real scenarios. This innovative method to enhance selectivity can avoid complex material synthesis and selection, and the highly specific nonenzymatic electrochemical biosensing platform paves the way for intelligent and precise point-of-care detection in long-term and is of low cost.