Computational Models for Central Nervous System Penetration

As the population of the elderly increases, there is a growing need for drugs to treat a variety of neurological diseases, such as Alzheimer's disease, Parkinson's disease, brain cancer, stroke, and infections in the central nervous system (CNS). Conversely, there is a need to identify brain penetration, and therefore potentially adverse events, from drugs acting on non-CNS targets. Late-stage clinical failures are costly in the drug discovery process, giving rise to the need for models to predict blood-brain barrier (BBB) penetration. In vivo and ex vivo models are expensive, time-consuming, and labor-intensive, giving rise to the development of in vitro and in silico models to aid in drug development early in the discovery process. Recent years have seen an increased emphasis on predictive computational models of CNS penetration. We review the progress in computational models of CNS penetration over the last five years. Computational models reported in the literature usually model the ratio of brain to blood levels for a molecule. These models can be broken down into logBB models, which attempt to predict a discrete value for the logarithm of the brain:blood ratio, and binary classifiers, which classify molecules as either brain-penetrant or non-penetrant according to an arbitrary cutoff. We also discuss whether the brain:blood ratio is an appropriate metric to use in predicting CNS penetration and the need for alternative endpoints that measure the information medicinal chemistry teams are actually interested in, such as the permeabilitysurface (PS) product and the fraction unbound (fu) in the brain.