关键词: ACE抑制剂, 手性, Vmedc, 苯基哌啶类, 反传神经网络, Fisher判别分析

Based on the interaction between different atomic types, Vmedc, a novel vector of molecular electronegative distance (Vmed) has been defined and generalized in order to further codify chemical structural information for chiral drugs. Some quantitative structure-activity relationships (QSAR) have been modeled by Vmedc for both 32 stereoisomers of perindoprilate as angiotensin-converting enzyme ACE inhibitors and 7 pairs of chiral N-alkylated 3-(3-hydroxyphenyl)-piperidines that bind σ-receptors. Stepwise linear regression analysis was made forward to the 32 stereoisomers with good modeling results: R＝0.913 (R2＝0.834, SD＝0.768, F＝33.875); Rcv＝0.877 (Rcv2＝0.769, SDcv＝0.906, Fcv＝22.473). Furthermore, average correlation coefficients (R) for random 60 groups with 23 training compounds for all the 32 ACE stereoisomers by backpropagation neural network (BPNN) were Rtr＝0.931 (Rtr2＝0.967) and Rcv＝0.918 (Rcv2＝0.842), except for four groups sampled unreasonably. Compared with literatures, Vmedc has also been applied to obtain good results for 14 samples with correlation coefficient being Rcv＝0.955 (Rcv2＝0.849). Through both Fisher’ linear discriminant analysis and BPNN, the 32 ACE stereoisomers were classified correctly into 88.89% active with one (#9) wrongly classified, 100.00% nonactive with no wrongly classified, and average classification of 96.87% globally. Good results obtained here were compared to those obtained with other chiral descriptors, when it was applied to the same 2 datasets, which shows that the Vmedc approach provides a powerful alternative QSAR technique for chiral compounds.

Key words: ACE inhibitor, chiral, Vmedc, N-alkylated-3-(3-hydroxyphenyl)piperidine, backpropagation neural network, linear discriminant analysis