Abstract: Structures of 67 aroma compounds of lemon peel were characterized through the molecular vertexes and relationships between molecular vertexes. Multiple linear regression (MLR) and partial least square regression (PLS) were adopted to construct two quantitative structure-retention relationship (QSPR) models. The estimation stability and predictability of the two models were strictly analyzed by both internal and external validations. Modeling multiple correlation coefficients (R) of the two models were 0.951 and 0.938,leave-One-Out (LOO) cross-validation (CV) correlation coefficients (R_CV) were 0.925 and 0.900,external forecasts multiple correlation coefficient (R_test) were 0.955 and 0.940,respectively. The results showed that the structural descriptors were superior in molecular structural representation. The estimated capacity,stability and predictability of the two models were good.

Key words: lemon peel, aroma components, structural descriptor, quantitative structure-retention relationship