Correlating sensory attributes to gas chromatography-mass spectrometry profiles and e-nose responses using partial least squares regression analysis.

Sensory scores for 15 attributes identified in soy sauce aroma by quantitative descriptive analysis were correlated with purge and trap gas chromatography-mass spectrometry (GC-MS) profiles and electronic nose (e-nose) responses using partial least squares (PLS) regression analysis. Highly predictive PLS models were obtained for every attribute based on whole GC-MS profiles. However, the predictability has been greatly improved in the models calculated from 20 selected peaks that showed higher contribution to each attribute in the first PLS analysis. Contrarily, except for alcoholic and fishy notes, predictability of PLS models calculated from e-nose responses was poor. The correlation between GC-MS profiles and e-nose responses was unsatisfactory due to high similarity in sensor responses.