QSAR studies of hallucinogenic phenylalkylamines by using neural network and support vector machine approaches

Quantitative structure-activity relationship (QSAR) studies based on a data set of 88 phenylalkylamines has been implemented. These chemicals used are among the most widely abused hallucinogens especially for young people. Because of the difficulty of assaying hallucinogenic activities, it is particularly important to develop predictive models. In this work, quantitative structure-activity relationships of phenylalkylamines were determined by using three methods, multiple liner regression (MLR), generalized regression neural network (GRNN), and support vector machine (SVM). Seven molecular descriptors, accounting for distributions of atomic charges, molecular orbital energy, molecular size and hydrophobic property were selected by stepwise regression method to build QSAR models. Comparison of the results obtained from the three models showed that the SVM and GRNN methods exhibited better performance than MLR method. SVM revealed better predictive performance comparing to the GRNN method. All the three methods should be useful to rapidly identify potential hallucinogenic phenylalkylamines.