Abstract

In this study, non-targeted 1H NMR fingerprinting was used in combination with multivariate statistical techniques for the classification of Italian sweet cherries based on their different geographical origins (Emilia Romagna and Puglia). As classification techniques, Soft Independent Modelling of Class Analogy (SIMCA), Partial Least Squares Discriminant Analysis (PLS-DA), and Linear Discriminant Analysis (LDA) were carried out and the results were compared. For LDA, before performing a refined selection of the number/combination of variables, two different strategies for a preliminary reduction of the variable number were tested. The best average recognition and CV prediction abilities (both 100.0%) were obtained for all the LDA models, although PLS-DA also showed remarkable performances (94.6%). All the statistical models were validated by observing the prediction abilities with respect to an external set of cherry samples. The best result (94.9%) was obtained with LDA by performing a best subset selection procedure on a set of 30 principal components previously selected by a stepwise decorrelation. The metabolites that mostly contributed to the classification performances of such LDA model, were found to be malate, glucose, fructose, glutamine and succinate.

Autore Pugliese

• Agostiano Angela , Longobardi Francesco , Catucci Lucia

Tutti gli autori

• AGOSTIANO A.;LONGOBARDI F.;CATUCCI L.

Titolo volume/Rivista

Non Disponibile

Anno di pubblicazione

2013

ISSN

0308-8146

ISBN

Non Disponibile

Numero di citazioni Wos

24

Ultimo Aggiornamento Citazioni

Non Disponibile

Numero di citazioni Scopus

24

Ultimo Aggiornamento Citazioni

Non Disponibile

Settori ERC

Non Disponibile

Codici ASJC

Non Disponibile