Application of GPA and PLSR in correlating sensory and chemical data sets

Abstract

This paper discusses the application of various multivariate statistical procedures to understand the relationship between sensory and instrumental flavor profiles. Ice cream with varying fat levels was used as the vehicle for the flavor compounds in the experiment. Chemical and sensory flavor profiles were obtained by modified dynamic headspace analysis and descriptive analysis, respectively. Chromatographic peak areas of flavor-volatiles were used as the variables for the chemical data set. Initially, principal component analysis and canonical variate analysis were performed separately on the chemical and sensory data sets to explore the structure of each set. Flavor volatiles were then further studied to investigate their impact on the sensory profile of ice cream using general procrustes analysis and partial least squares regression analysis. The results from the two statistical analysis methods are compared and discussed. Additionally, the effect of log-transformation of chemical data on the overall chemical–sensory relationship was evaluated within each statistical method.