A multi-technique data fusion approach applied to a case study on Tonda Gentile Trilobata from Piedmont, Italy

The increasing attention to food traceability calls for robust and sensitive analytical methodologies. Here we applied multivariate statistical analysis on data obtained from different analytical techniques typically used to determine food authenticity, namely nuclear magnetic resonance (1H-NMR), liquid chromatography high-resolution mass spectrometry (LC-HRMS), and bulk stable isotope analysis (BSIA). The data were firstly analysed as independent data sets and then the compatible data were merged using a multi-omics data fusion approach. As a case study, Tonda Gentile Trilobata hazelnut from Piemonte (known as Nocciola Piemonte PGI) was used, considering different origins and cultivars collected for two consecutive years. A first exploration of data from each technique revealed a strong temporal component therefore the information related to origin and cultivar was evaluated on data from each year separately; the three techniques highlighted differences between origins, while only 1H-NMR and LC-HRMS could discriminate cultivars. The 1H-NMR and the LC-HRMS datasets were then merged using Data Integration Analysis for Biomarker discovery using Latent variable approaches for Omics studies (DIABLO) framework, a supervised multivariate method for multi-omics integration, designed to identify correlated variables across datasets and maximize class discrimination. The merged data were adequately classified for the geographical origin and cultivar with minimum error rate while the features across the two datasets recorded similar up or down-modulation. Data fusion also confirmed the hierarchically stronger effect of annual variability in agreement with the outcome of individual analytical approaches. With this work, we show that data fusion increases robustness and enhances the extracted information by leveraging the strengths of each analytical technique.