Multi-component metabolite electrochemical detection and analysis based on machine learning

Abstract

Metabolic molecules are highly correlated with various physiological indicators and diseases, so it is particularly important to monitor the levels of multiple metabolites in the body. Due to the similar electrochemical properties of uric acid (UA), dopamine (DA), and ascorbic acid (AA), multi-component detection of these substances is challenging. When establishing relationships between electrochemical characteristics and concentrations of the respective components, there will be issues such as overlapping peaks and other difficulties. In order to accurately identify the components and determine their concentration in the detection solution, we designed a multi-component detection experiment for AA, UA, and DA. After obtaining the detection results, we applied curve smoothing and feature extraction to construct classification and regression machine learning models. The ANN model achieved the highest accuracy of 94.06% among the five classification models evaluated. Regression models were built using RF and XGBoost, with the best performing XGBoost model achieving an average R-squared prediction of 96.2%. With high component discrimination and prediction accuracy, these models ensure user-friendliness and support qualitative and quantitative analysis of multi-component solutions.