Multivariate Modeling of Sensory and Chemical Data to Understand Staling in Light Beer

Abstract

To identify beer chemistry changes during staling in light beer, an attempt was made to model analytical chemistry data with sensory data using multivariate statistical methods on samples subjected to various time-temperature treatments. Standard beer analyses were completed on these light beer samples in addition to using an in-house gas chromatographic method to profile carbonyl development during staling. Trained sensory panelists evaluated these packaged beer samples following standard descriptive methods and rated a total of 31 attributes, of which 12 flavor and aroma attributes related to staling were selected for modeling. Projection to latent structures (PLS) algorithms were employed on the data to model the 48 known and 41 unknown carbonyl compounds to these 12 mean sensory scores. The known compounds found to be significant indicators of light beer staling from the PLS models were 2-furfural, acetal, 5-hydroxymethyl-2-furfural, t,t-2,4-hexadienal, dihydro-5-pentyl-2(3H)-furanone (γ-nonalactone), t,t-2,4-decadienal, and 1-heptanol. Conversely, compounds that loaded negatively in the PLS models for staling were furfuryl acetate, ethyl hexanoate, 1-heptyl acetate, and ethyl octanoate (i.e., these esters decrease during staling). In addition, several unknown peaks loaded both positively and negatively with beer staling in the PLS model. This PLS methodology may prove to be a useful tool for understanding which compounds drive staling in certain beers. In our models, light beer staling displayed a dual mechanism in which known unpleasant stale compounds increased at the same time that known pleasant fruity esters decreased.