Reference samples guide variable selection for correlation of wine sensory and volatile profiling data

Abstract The relationship between wine flavour and wine volatile composition is well recognised, however with thousands of compounds in wine the exact nature of individual contributions may be hard to determine due to synergistic and masking effects. Untargeted chemical analyses coupled with descriptive sensory and partial least squares regression modelling can help unravel interactions to identify groups of compounds that contribute to sensory properties. Variable selection is often applied prior to modelling to eliminate irrelevant variables. In this study, sensory references used to train the sensory panel were chemically analysed and employed to reduce the number of variables used to construct the models. This novel variable selection approach was compared against the inclusion of all variables and the most commonly applied variable selection method – analysis of variance. Models constructed from variables present in sensory references performed similarly to other models and identified interesting groups of compounds to investigate further.