Improvement of the Prediction Power of the CoMFA and CoMSIA Models on Histamine H3 Antagonists by Different Variable Selection Methods

The aim of this study is to enhance the predictivity power of CoMFA and CoMSIA models by means of different variable selection algorithms. The genetic algorithm (GA), successive projection algorithm (SPA), stepwise multiple linear regression (SW-MLR), and the enhanced replacement method (ERM) were used and tested as variable selection algorithms. Then, the selected variables were used to generate a simple and predictive model by the multilinear regression algorithm. A set of 74 histamine H₃ antagonists were split into 40 compounds as a training set, and 17 compounds as a test set, by the Kennard-Stone algorithm. Before splitting the data, 17 compounds were randomly selected from the pool of the whole data set as an evaluation set without any supervision, pretreatment, or visual inspection. Among applied variable selection algorithms, ERM had noticeable improvement on the statistical parameters. The r² values of training, test, and evaluation sets for the ERM-MLR model using CoMFA fields were 0.9560, 0.8630, and 0.8460 and using the CoMSIA fields were 0.9800, 0.8521, and 0.9080, respectively. In this study, the principles of organization for economic cooperation and development (OECD) for regulatory acceptability of QSARs are considered.