Prediction of relative sensitivity of the olfactory and nasal trigeminal chemosensory systems for a series of the volatile organic compounds based on local lazy regression method

Quantitative structure–activity relationship (QSAR) models were successfully developed for predicting the relative sensitivities odor detection thresholds (ODTs) and nasal pungency thresholds (NPTs) for the olfaction and nasal trigeminal chemosensory systems of a set of volatile organic compounds (VOCs). The best multi-linear regression (BMLR) method was used to select the most important molecular descriptors and build a linear regression model. The methods support vector machine (SVM) and local lazy regression (LLR) were also used to build regression models. By comparing the results of these methods for the test set of ODTs and NPTs, the LLR model gave better results for the VOCs with the coefficient of determination R2 (0.9171, 0.9609, respectively) and root mean square error (RMSE) (0.3861, 0.2152, respectively). At the same time, this study identified some important structural information which was strongly correlated to the relative sensitivities of these VOCs. Such information can be used to select and manufacture chemical sensors. As it could predict accurately the relative sensitivities of the olfaction and nasal chemesthesis, the LLR method is a promising approach for QSAR modeling, and it also could be used to model the other similar chemical sensors.