In Silico Prediction of Blood Brain Barrier Permeability

Abstract

The ability of penetration of the blood-brain barrier is an important property for the development of Central Nervous System drugs, which is commonly expressed by logBB (logBB = log(Cbrain/Cblood). In this work, a support vector machine was used to build quantitative models of blood brain barrier permeability. Molecular descriptors for 182 compounds were calculated by ADRIANA.Code and 12 descriptors were selected using the automatic variable selection function in Weka. Based on two common physicochemical descriptors (xlogP and Topological Polar Surface Area (TPSA)) and 10 2D property autocorrelation descriptors on atom pair properties, an SVM regression model was built. The built model was validated by an external test set. The reliable predictions of the test set demonstrate that this model performs well and can be used for estimation of logBB values for drug and drug-like molecules.