Integrating volatilome, primary and specialized metabolome by data fusion techniques: a comprehensive evaluation of hazelnuts quality

Data fusion is the process of integrating information from multiple sources to produce more accurate and comprehensive data giving access to higher level information, i.e., understanding of complex phenomena. The practice has become increasingly important in recent years as the volume of data being generated by various sources continues to grow, as in the case of “omics” analytical approaches [1]. The process of data fusion involves several stages, including data acquisition, data cleaning and pre-processing, variable selection, and feature extraction, with each stage being critical in ensuring the accuracy and completeness of the fused data [2]. In this study, the data related to targeted and untargeted features generated by different analytical techniques (i.e., comprehensive two-dimensional gas chromatography coupled to time-of-flight mass spectrometry - GC×GC-TOFMS, liquid chromatography coupled to high-resolution mass spectrometry – LC-HRMS etc.) were fused to reach a comprehensive understanding of hazelnuts quality along their shelf-life. Volatiles, primary metabolites, and selected specialized metabolites, collected as known and unknown features were investigated on a unique sample set that included different cultivars/geographical origins, post-harvest practices, and storage time/conditions. The comprehensive dataset revealed the susceptibility of the various metabolites to shelf-life related phenomena, with the volatilome as more informative for storage impact and sensory quality, followed by specialized metabolites (mainly phenols and polyphenols) whose patterns are quite stable and more strongly related to cultivar and geographical origin. Last but not least, primary metabolites react to post-harvest practices influencing the aroma potential of raw fruits. Regarding the effectiveness of the models, data fusion not only improves global classification/prediction capabilities but also lowers the level of uncertainty associated with each individual result and improves outlier detection in prediction.