ADME Evaluation in Drug Discovery. 3. Modeling Blood-Brain Barrier Partitioning Using Simple Molecular Descriptors

In this paper, QSPR models were developed for in vivo blood-brain partitioning data (logBB) of a large
data set consisting of 115 diverse organic compounds. The best model is based on three descriptors:  n-octanol/water partition coefficient calculated using the SLOGP approach, logP; high-charged polar surface areas
based on the Gasteiger partial charges, HCPSA, and the excessive molecular weight larger than 360, MW360.
The model bears good statistical significance, n = 78, r = 0.88, q = 0.86, s = 0.36, F = 81.5. The actual
prediction potential of the model was validated through two external validation sets of 37 diverse compounds.
The predicted results demonstrate that the model bears better prediction potential than many other models
and can be used for logBB estimations for drug and drug-like molecules. Comparison of several logP
calculation approaches suggests that logP calculated by SLOGP can be used as a significant descriptor for
the prediction of molecular transport properties because SLOGP gives the most similar results with CLOGP.
The QSPR model indicates that larger polar surface areas have a more negative contribution to logBB, but
the absolute partial charges on the atoms surrounded by the polar surfaces should be larger than 0.10|e|.
Meanwhile, tight junction membranes limit the size of hydrophilic molecules that can cross the membrane
with a molecular weight of approximately 360, because when a molecule's weight is larger than 360 it
shows a negative contribution to logBB. The computations of molecular surface, partial charges, logP, and
logBB have been accomplished using a program called Drug-BB. Moreover, to improve the efficiency of
the computations of logP, we made an extensive reparametrization of SLOGP, and the newly developed
SLOGP model is only based on simple atomic addition. Further, we developed a set of parameters to calculate
the topological polar surface area (TPSA), thus the high-charged topological polar surface area (HCTPSA)
could be estimated from the 2D connection information of a molecule. Adopting the new strategies, the
estimations of logP, HCTPSA, and logBB are only based on the topological structure of a molecule and
therefore, can be used for fast screening of virtual libraries having millions of molecules.