Soroush Ganji - Department of Chemistry, Pharmaceutical Sciences Branch, Islamic Azad University, Tehran, Iran
Shahin Ahmadi - Department of Chemistry, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran

The data set splitting and variable selection is two important stages in QSPR modeling. The data set splitting of QSPR models of a group of cholinesterase inhibitors based on random division and self-organizing maps (SOM) was compared. Then GA-MLR and stepwise multiple linear regression as two methods for variable selection. A set of N-aryl substituted derivatives (44 amides and 44 imides) which can inhibit the catalytic activities of the cholinesterase enzymes were selected from literature and a large number of theoretical descriptors was calculated for each molecule using Dragon software. The random sampling of the training set was performed and the remaining molecules were used as external validation set and then SOM data splitting was performed on data set. Each time, the most appropriate QSPR model was produced by GA-MLR and stepwise multiple linear regression. The external validation statistics were reported for each model as a basis for the final comparison. As the results, we found that SOM division and GA-MLR method can be employed as reliable methods to develop a predictive QSPR models

QSPR , S-MLR , GA-MLR , Self-organizing maps