Artificial Intelligence (AI) smelling based on GC×GC: a key-tool to make a step forward in food quality measurements

The volatile fraction of food, also referred to as food volatilome, encrypts information on quality and authenticity of a specific product. In this contribution, the industrial quality of hazelnuts (Corylus avellana L.), is exploited to show the potentials of Artificial Intelligence (AI) smelling based on sensomics and realized on fully automated GC×GC platforms. Hazelnut volatilome encrypts information on cultivar/geographical origin, post-harvest treatments, bacteria/molds contamination, oxidative stability, and overall sensory perception. By comprehensive two-dimensional gas chromatography (GC×GC), untargeted/targeted features patterns can be effectively mapped (i.e., fingerprinting) across the volatile fingerprint of many samples to answer different questions. However, chromatographic fingerprinting consistency, is hardly compromised in long-term studies by random errors generated by chromatographic issues (e.g., retention times shifts) or detector response fluctuations, generally identified as analytical batch effects. Simpler, yet more robust, approaches are therefore preferred in long-term monitoring, usually adopting 1D-GC and targeted quantifications. In this study, the AI smelling machine approach is implemented on a flow modulated GC×GC-MS/FID platform, improvements related to extended accurate quantification, via multiple headspace solid phase microextraction (MHS-SPME), of over 40 analytes are discussed. Within targeted compounds, key-aromas, potent odorants, and spoilage markers are included. Quantitative “AI” perceptive maps, based on odor activity values (OAVs), enable consistent samples clustering based on the overall smell, while guaranteeing results transferability over time. Of particular interest is the accurate detection of spoilage smells, which makes this technique suitable for industrial quality control purposes.