A novel set of structural descriptors with ten elements, called the hydrogen-association classified molecular electrongativity-distance vector (H-MEDV), has been developed according to classification of hydrogen connection for various local atoms in the molecules under examination. The H-MEDV vector was then used to describe the chemical structure of 22 cyclic ureas and reasonable molecular modeling results were achieved with a very high correlation coefficient (R = 0.971) by a multiple linear regression (MLR). Another predictive model was constructed with a quite high correlation coefficient (R = 0.938) by selecting five parameters from all ten elements mentioned above in the H-MEDV vector of the former model through a stepwise multiple regression (SMR). The performance of the reduced five-parameter model was tested through cross-validation by the leave-one-out procedure (LOO) and satisfactory results were obtained with a fairly resemble correlation coefficient (R = 0.908), which shows the estimation stability and predictive ability of the model based on the H-MEDV descriptors.

Key words: cyclic ureas, AIDS, QUANTITATIVE STRUCTURE ACTIVITY RELATIONSHIP