Relating sensory properties of tea aroma to gas chromatographic data by chemometric calibration methods

Seven sensory properties in tea aromas such as fresh floral, sweet floral, citrus, sweet fruity, fresh green, resinous and roasted were statistically correlated to the GC profiles of volatile flavour components by multivariate calibration methods. Stepwise multiple linear regression analysis (MLR), principal component regression (PCR) and partial least squares (PLS) regression analyses were comparatively applied to the sensory scores and the 77 GC peaks. The logarithmic transformation of GC data considerably improved the coefficients of determination (R(2)) in, every calibration method. MLR, PCR and PLS succeeded in calculating highly predictable regression models for most aroma properties except for resinous and roasted. Among 77 volatile components, Linalool and jasmine lactone, 2-phenylethanol and jasmine lactone, and 2-phenylethanol highly contributed for the prediction of fresh floral, sweet floral and sweet fruity, respectively. Linalool oxide (II) heavily weighted on citrus but the sign for linalool oxide (I) became negative due to multicollinearity existing between the two isomers. The R(2)s for roasted and resinous were comparatively low in the three methods. However, the positive contribution of two pyrrole derivatives for roasted agreed well with the practical sense in flavour chemistry.