Predicting penetration across the blood-brain barrier from simple descriptors and fragmentation schemes

The ability to cross the blood brain barrier (BBB), sometimes expressed as BBB+ and BBB-, is a very important property in drug design. Several computational methods have been employed for the prediction of BBB-penetrating (BBB+) and nonpenetrating (BBB-) compounds with overall accuracies from 75 to 97%. However, most of these models use a large number of descriptors (67-199), and it is not easy to implement the models in order to predict values of BBB+/-. In this work, 19 simple molecular descriptors calculated from Algorithm Builder and fragmentation schemes were used for the analysis of 1593 BBB+/- data. The results show that hydrogen-bonding properties of compounds play a very important role in modeling BBB penetration. Several BBB models based on hydrogen-bonding properties, such as Abraham descriptors, polar surface area (PSA), and number of hydrogen bonding donors and acceptors, have been built using binomial-PLS analysis. The results show that the overall classification accuracy for a training set is over 90%, and overall prediction accuracy for a test set is over 95%.