Preventing mislabeling of organic white button mushrooms (Agaricus bisporus) combining NMR-based foodomics, statistical, and machine learning approach

Abstract

Organic foods are among the most susceptible to fraud and mislabeling since the differentiation between organic and conventionally grown food relies on a paper-trail-based system. This study aimed to develop a differentiation model that combines nuclear magnetic resonance (NMR), statistical approach (principal component analysis – PCA and partial least square discriminant analysis – PLS-DA), and classification artificial neural network (cANN). The model was tested for hydrophilic and lipophilic extracts of Agaricus bisporus. As linear techniques, the PCA and PLS-DA analyses and cANN as a non-linear classification tool successfully discriminated organic from conventional samples regarding their NMR data. PLS-DA revealed higher similarity among the hydrophilic samples within the organic class and among the lipophilic samples within the conventional class. Both applied approaches demonstrated high statistical quality, but a higher level of classification confidence in the case of lipophilic extracts. The metabolites responsible for discrimination and observed (dis)similarities between classes were considered according to cultivation specificities.