Untargeted metabolomics and machine learning unveil quality and authenticity interactions in grated Parmigiano Reggiano PDO cheese.

The chemical composition of Parmigiano Reggiano (PR) hard cheese can be significantly affected by different factors across the dairy supply chain, including ripening, altimetric zone, and rind inclusion levels in grated hard cheeses. The present study proposes an untargeted metabolomics approach combined with machine learning chemometrics to evaluate the combined effect of these three critical parameters. Specifically, ripening was found to exert a pivotal role in defining the signature of PR cheeses, with amino acids and lipid derivatives that exhibited their role as key discriminant compounds. In parallel, a random forest classifier was used to predict the rind inclusion levels (> 18%) in grated cheeses and to authenticate the specific effect of altimetry dairy production, achieving a high prediction ability in both model performances (i.e., ∼60% and > 90%, respectively). Overall, these results open a novel perspective to identifying quality and authenticity markers metabolites in cheese. • Ripening, altimetry, and rind inclusion affected grated Parmigiano Reggiano quality. • Untargeted metabolomics and machine learning were robust authenticity tools. • Amino acid and lipid derivatives were confirmed as key ripening markers. • Random Forest successfully predicted forbidden (> 18%) rind inclusion levels. • A 90.63% predictability was achieved in assessing mountain dairy productions. [ABSTRACT FROM AUTHOR]