Predicting Blood−Brain Barrier Permeation from Three-Dimensional Molecular Structure

TLDR

Predicting blood‑brain barrier permeation is challenging, and experimental determination for many candidates is infeasible, making computational models desirable. The study aims to demonstrate that descriptors derived from 3D molecular fields can estimate BBB permeation and to develop a simple external prediction model. Using VolSurf, 3D fields are transformed into descriptors that are linked to experimental permeation through a discriminant partial least squares procedure. The model accurately predicts more than 90 % of BBB permeation data, quantifies favorable and unfavorable physicochemical and structural contributions, is fully automated and fast, and provides a valuable tool for virtual screening and drug design.

Abstract

Predicting blood-brain barrier (BBB) permeation remains a challenge in drug design. Since it is impossible to determine experimentally the BBB partitioning of large numbers of preclinical candidates, alternative evaluation methods based on computerized models are desirable. The present study was conducted to demonstrate the value of descriptors derived from 3D molecular fields in estimating the BBB permeation of a large set of compounds and to produce a simple mathematical model suitable for external prediction. The method used (VolSurf) transforms 3D fields into descriptors and correlates them to the experimental permeation by a discriminant partial least squares procedure. The model obtained here correctly predicts more than 90% of the BBB permeation data. By quantifying the favorable and unfavorable contributions of physicochemical and structural properties, it also offers valuable insights for drug design, pharmacological profiling, and screening. The computational procedure is fully automated and quite fast. The method thus appears as a valuable new tool in virtual screening where selection or prioritization of candidates is required from large collections of compounds.