Volatile-Based Prediction of Sauvignon Blanc Quality Gradings with Static Headspace-Gas Chromatography-Ion Mobility Spectrometry (SHS-GC-IMS) and Interpretable Machine Learning Techniques

The analytical scope of static headspace-gas chromatography-ion mobility spectrometry (SHS-GC-IMS) was applied to wine aroma analysis for the first time. The method parameters were first fine-tuned to achieve optimal analytical results, before the method stability was demonstrated, in terms of repeatability and reproducibility. Succinct qualitative identification of compounds was also realized, with the identification of several volatiles that have seldom been described previously in Sauvignon Blanc wine, such as methyl acetate, ethyl formate, and amyl acetate. Using the SHS-GC-IMS data in an untargeted approach, computer modeling of large datasets was applied to link aroma chemistry via prediction models to wine sensory quality gradings. Six machine learning models were compared, and artificial neural network (ANN) returned the most promising performance with a prediction accuracy of 95.4%. Despite its inherent complexity, the ANN model offered intriguing insights on the influential volatiles that correlated well with higher and lower sensory gradings. These findings could, in the future, guide winemakers in establishing wine quality, particularly during blending operations prior to bottling.

Automated summary

In this study, SHS-GC-IMS technology was applied to wine aroma analysis for the first time, achieving optimal analytical results through parameter fine-tuning and demonstrating method stability. Several previously undescribed volatile compounds were identified in Sauvignon Blanc wine. Using machine learning models, particularly artificial neural network, large datasets of aroma chemistry data were linked to wine sensory quality gradings with a high prediction accuracy of 95.4%.