In silico ADME modelling: prediction models for blood-brain barrier permeation using a systematic variable selection method

摘要

Quantitative Structure-Property Relationship models (QSPR) based on in vivo blood-brain permeation data (logBB) of 88 diverse compounds, 324 descriptors and a systematic variable selection method, namely 'Variable Selection and Modeling method based on the prediction (VSMP)', are reported. Of all the models developed using VSMP, the best three-descriptors model is based on Atomic type E-state index (SsssN), AlogP98 and Van der Waal's surface area (r = 0.8425, q = 0.8239, F = 68.49 and SE = 0.4165); the best four-descriptors model is based on Kappa shape index of order 1, Atomic type E-state index (SsssN), Atomic level based AI topological descriptor (AIssssC) and AlogP98 (r = 0.8638, q = 0.8472, F = 60.982 and SE = 0.3919). The performance of the models on three test sets taken from the literature is illustrated and compared with the results from other reported computational approaches. Test set Ⅲ constitutes 91 compounds from the literature with known qualitative BBB indication and is used for virtual screening studies. The success rate of the reported models is 82% in the case of BBB + compounds and a similar success rate is observed with BBB - compounds. Finally, as the models reported herein are based on computed properties, they appear as a valuable tool in virtual screening, where selection and prioritization of candidates is required.