In Silico Modeling on ADME Properties of Natural Products: Classification Models for Blood-Brain Barrier Permeability, Its Application to Traditional Chinese Medicine and in Vitro Experimental Validation

Xiuqing Zhang,Ting Liu,Xiaohui Fan,Ni Ai

DOI: https://doi.org/10.1016/j.jmgm.2017.05.021

2017-01-01

Abstract:In silico modeling of blood-brain barrier (BBB) permeability plays an important role in early discovery of central nervous system (CNS) drugs due to its high-throughput and cost-effectiveness. Natural products (NP) have demonstrated considerable therapeutic efficacy against several CNS diseases. However, BBB permeation property of NP is scarcely evaluated both experimentally and computationally. It is well accepted that significant difference in chemical spaces exists between NP and synthetic drugs, which calls into doubt on suitability of available synthetic chemical based BBB permeability models for the evaluation of NP. Herein poor discriminative performance on BBB permeability of NP are first confirmed using internal constructed and previously published drug-derived computational models, which warrants the need for NP-oriented modeling. Then a quantitative structure property relationship (QSPR) study on a NP dataset was carried out using four different machine learning methods including support vector machine, random forest, Naive Bayes and probabilistic neural network with 67 selected features. The final consensus model was obtained with approximate 90% overall accuracy for the cross-validation study, which is further taken to predict passive BBB permeability of a large dataset consisting of over 10,000 compounds from traditional Chinese medicine (TCM). For 32 selected TCM molecules, their predicted BBB permeability were evaluated by in vitro parallel artificial membrane permeability assay and overall accuracy for in vitro experimental validation is around 81%. Interestingly, our in silico model successfully predicted different BBB permeation potentials of parent molecules and their known in vivo metabolites. Finally, we found that the lipophilicity, the number of hydrogen bonds and molecular polarity were important molecular determinants for BBB permeability of NP. Our results suggest that the consensus model proposed in current work is a reliable tool for prioritizing potential CNS active NP across the BBB, which would accelerate their development and provide more understanding on their mechanisms, especially those with pharmacologically active metabolites. (C) 2017 Elsevier Inc. All rights reserved.