A Predictive Model for Blood-Brain Barrier Penetration #
BioChem Press, FU Xu-chun, Song Zhifang, Wenquan Liang · 2005
Motivation. It is important to determine whether a candidate molecule is capable of penetrating the blood–brain barrier in drug discovery and development. The aim of this paper is to develop a predictive model for blood– brain barrier penetration only using two simple descriptors, molecular volume (V) and polar surface area (PSA, defined as the sum of the van der Waals surface areas of oxygen atoms, nitrogen atoms, and attached hydrogen atoms in a molecule). Method. A data set of 100 compounds, which was studied by other research groups, is divided into a training set of 61 compounds and two test sets (14 and 25 compounds). Molecular volumes and polar surface areas are obtained from the molecular conformations optimized using the semiempirical self–consistent field molecular orbital calculation AM1 method. The model to predict blood–brain barrier penetration from molecular volume and polar surface area is derived on the training set using the stepwise multiple regression analysis and then cross–validated using leave–one–out procedure and tested on the external prediction. Results. A logBB model is developed using the training set of 61 compounds (4 compounds are excluded as outliers): logBB = –16.79 (±3.28) V 2 + 11.24 (±2.06) V – 2.249 (±0.161) PSA – 0.6583 (±0.2326) (n = 57, r 2 = 0.832, q 2 = 0.804, s = 0.329, F = 87.2), where logBB is the logarithm of the ratio of the steady–state concentration of a compound in the brain to in the blood, n is the number of compounds, r is the correlation coefficient, q is the cross validation coefficient, s is the standard deviation, F is the Fisher F–statistic. The model is validated through two external test sets (14 compounds and 25 compounds). The root mean squared errors (RMSE) are 0.599 for test set 1 of 14 compounds and 0.551 for test set 2 of 25 compounds. The simple model performs as well as other logBB models developed using the same data set but different descriptors. Conclusions. The model derived in this paper for the prediction of BBB penetration shows a good predictive power. It shows that the hydrogen–bonding potential, lipophilicity, and molecular size are important factors to affect BBB penetration. The model is one of the simplest logBB models and suitable for the rapid prediction of the BBB penetration for a wide range of drug candidates.