Unconventional Molecular Descriptors for Blood-Brain Barrier Permeability Prediction: A Unified Model and Its...

Predicting blood-brain barrier (BBB) permeability remains a central problem in central nervous system (CNS) drug design. Conventional rules such as Lipinski's rule of five and related CNS guidelines describe physicochemical boundaries but do not guarantee permeability.

We examined how a set of molecular descriptors-including membrane cross-sectional area, collision cross-section, P-glycoprotein (P-gp) net flux, desolvation cost, chameleonicity, 3D polar surface area, dipole moment/polarizability, LUMO energy, substructural synergy, and lateral bilayer pressure-can be integrated into a single probabilistic framework.

A curated set of 24 drugs (17 BBB-permeable, 7 BBB-poor) was compiled from the published literature. Descriptor values were estimated from public structural data, published binding/efflux data, and validated biophysical relationships; the relative membrane-partition term was calculated from the lateral bilayer pressure model.

Membrane cross-sectional area (A_D) and P-gp net flux discriminated BBB-positive from BBB-negative compounds most clearly. Desolation cost distinguished morphine from heroin and codeine, and lateral bilayer pressure provided a physical rationale for the exponential size dependence of membrane partitioning. Caffeine, ethanol, and nicotine, all with molecular weight below 200 Da, are examples of small molecules that can cross by paracellular or small-molecule diffusion, so their permeability does not rely solely on lipid-membrane partitioning.

BBB permeability can be framed as the product of three gated probabilities: desolation, membrane partition, and net transmembrane flux. The model is a heuristic synthesis intended to guide hypothesis generation; the small, literature-derived dataset and the illustrative nature of the calculations mean that all therapeutic suggestions are strictly indicative.