SUPPORTING THE OFFICIAL SENSORY EVALUATION OF VIRGIN OLIVE OILS WITH INNOVATIVE STRATEGIES BASED ON VOLATILE COMPOUNDS ANALYSIS: THE ROLE OF HS-GC-IMS AND MACHINE LEARNING

Olive oil companies, as well as public and private laboratories, need to rapidly assess the quality and purity of olive oil samples. Such high-throughput screening methods can successfully contribute to improve the overall efficiency of official controls. Sensory analysis of virgin olive oils (performed through a well-established and effective method, known as Panel test) plays a pivotal role in quality control but suffers from a few drawbacks: it requires a panel of trained experts who need to meet in the same equipped room, a limited number of samples can be evaluated per day, and assessors may be affected by sensory fatigue. In this framework, a non-destructive screening method based on gas chromatography coupled with ion mobility spectrometry (HS-GC-IMS) for the analysis of volatile compounds, with the application of machine learning and chemometric techniques, can represent a useful tool to support the Panel test. It may be exploited as a routine method both through targeted and untargeted approaches. A set of around 100 commercial virgin olive oils was sensory assessed by five different panels and robustly classified, applying a decision tree based on the panels’ agreement, previously set-up and published. Moreover, the volatile fraction of these oils was analysed in five laboratories by HS-GC-IMS with a targeted approach in which fifteen relevant molecules, as monomers and dimers, were considered to build separate predictive models. The commercial category of each oil was predicted based on volatile compounds profiles using PLS-DA models, with a focus on the borderline samples between the extra virgin and virgin, as they are particularly relevant in the real quality control scenario. The herein-described inter-laboratory study showed satisfactory results: most of the samples were correctly classified by the models in the same quality grade with respect to robust sensory classification. For the untargeted approach, 198 commercial virgin olive oils were sensory assessed by a professional panel. Then, the same oils were analysed by HS-GC-IMS and the data were extracted using the CSV mode from the .MEA file. Pre-processing is based on the transformation of the CSV file (1818 rows x 4500 columns wide) to an image. The images, each for every sample, were processed using the machine learning approach consisting of training, testing (to determine the performance index), and validation of the model. Different boosting and bagging techniques were compared with traditional statistics, both on classification and regression modes. All computations were executed in Python 3 environment using Scikit-learn 1.5.0 and OpenCV 4.10.0 libraries. The herein-discussed findings, together with previous investigations that will be briefly presented, support the use of HS-GC-IMS with machine learning and chemometrics as a useful tool to support the Panel test in the olive oil quality control. Keywords: extra virgin olive oil, volatile compounds, sensory analysis, HS-GC-IMS, machine learning Acknowledgement: Authors acknowledge the involved sensory panels, olive oil companies and laboratories, and Federolio for funding. The research activity is also funded by the project “AI applied to the optimization of production processes for the quality and sustainability of the production of edible and extra virgin olive oils; (OLEUM SPEC) financed by Next Generation EU - MEASURE M4C2 I2.3 PNRR. Dr. Enrico Casadei’s research activity is financed within the project funded under the National Recovery and Resilience Plan (NRRP) - NextGenerationEU “ON Foods - Research and innovation network on food and nutrition Sustainability, Safety and Security - Working ON Foods”.