Machine learning-assisted colorimetric sensor array for rapid identification of adulterated Panax notoginseng powder

Abstract

Panax notoginseng (PN) is a popular functional food worldwide, yet adulterated PN powders are prevalent in the commercial market. Herein, a method combining a low-cost colorimetric sensor array and machine learning is proposed to rapidly identify and quantify adulterated PN powder. We constructed a sensitive indicator displacement assay (IDA) sensor array by innovatively using Job's plot to optimize sensor units. The array was used to identify adulterated PN powders after its discriminatory ability was evaluated by amino acid analysis. Among the four machine learning models, the support vector machine (SVM) models achieved the highest accuracy, with above 98.3% for authenticity and 99.0% for adulteration types. The blending percentages of four PN adulterants were further analyzed quantitatively using support vector machine regression (SVR) models with good prediction ability (R > 0.93). Finally, our sensor array method was applied to identify commercially available PN powders with satisfactory results. This study offers a low-cost new method for the rapid identification of PN powder, contributing to quality control of powdered food.