Abstract: Totally 457 physicochemical variables of 20 kinds of natural amino acids were classified according to hydrophobic properties, electronic characteristics, hydrogen bonds contributions and steric properties, and analyzed by principal component analysis (PCA). A new amino acid structure descriptor, SVHEHS, was achieved. The descriptor was used to characterize the structures of a series of ACE inhibitory dipeptides and bitter dipeptides. A multiple linear regression (MLR) model was established on the basis of bioactivity. The correlation coefficients (R2), cross-validation correlation coefficients (Q2LOO) and root mean square errors (RMSE) for the models established for ACE inhibitory dipeptides and bitter dipeptides were 0.936, 0.854,   0.259, and 0.949, 0.886, 0.136, respectively. External validation was also conducted to validate the prediction capability of the models. The results showed that the models obtained with the descriptor had good fitting and prediction capability and consequently could be used for QSAR studies.

Key words: amino acid descriptor, quantitative structure activity relationship, principal component analysis, multiple linear regression