ИСТИНА |
Войти в систему Регистрация |
|
ИПМех РАН |
||
The ability to penetrate the blood-brain barrier is among the most important ADME parameters of drugs that control their bioavailability and play an important role in determining their activity. For the CNS-targeted drugs the permeability should be enhanced while for other drugs it should be minimized to avoid possible side effects. Substantial work has been devoted to the modelling and prediction of this property. However, the applicability and usefulness of available models is often diminished due to limited data sets, inaccurate data, and/or insufficiently validated modelling approaches. We have attempted to build generally applicable predictive models for the blood-brain barrier permeability (LogBB) of diverse drugs and drug-like compounds. The fragmental descriptors of up to 6 atoms were used in the conjunction with back-propagation neural networks (BPNN) in the framework of NASAWIN software [1]. The descriptor subset for BPNN modelling was preselected using fast stepwise multiple linear regression (FSMLR). The model predictivity was assessed by 5x4-fold double cross-validation procedure. We have compiled, to our knowledge, the most complete data set based on the open quantitative LogBB data (significantly extended over the largest previously published sets). The values were verified and errors corrected against the original publications. On the other hand, inorganic and small organic molecules irrelevant to medicinal chemistry were excluded. The final dataset contained 510 diverse organic compounds. The optimal model has Q2 value (double cross-validation) of 0.79 and RMSE of 0.34. The model is implemented in a convenient predictor software. For an additional check we used the recently published dataset of 2053 compounds [2] containing estimated qualitative values (BBB+ / BBB–). The predicted LogBB values were converted to the qualitative scale using the cut-off value LogBB = –1. This procedure gives total accuracy of 0.80, sensitivity of 0.91, specificity of 0.42, and precision of 0.84. In other words, more than 80% of BBB+ compounds in this independent validation set were identified correctly. [1] Baskin I.I., Halberstam N.M., Artemenko N.V., Palyulin V.A., Zefirov N.S., NASAWIN – a universal software for QSPR/QSAR studies. In: EuroQSAR 2002. Designing Drugs and Crop Protectants: Processes, Problems and Solutions, M. Ford, D. Livingstone, J. Dearden, and H. Waterbeemd, eds., Blackwell, Malden, 2003, pp. 260-263. [2] Martins I.F., Teixeira A.L., Pinheiro L., Falcao A.O., A Bayesian approach to in silico blood-brain barrier penetration modeling, J. Chem. Inf. Mod., 2012, 52 (6), 1686-1697.