Computational Model To Predict the Fraction of Unbound Drug in the Brain

Knowing the value of the unbound drug fraction in the brain (f(u,brain)) is essential in estimating its effects and toxicity on the central nervous system (CNS); however, no model to predict f(u,brain) without experimental procedures is publicly available. In this study, we collected 253 measurements from the literature and an open database and built in silico models to predict f(u,brain) using only freely available software. By selecting appropriate descriptors, training, and evaluation, our model showed an acceptable performance on a test data set (R-2 = 0.630, percentage of compounds predicted within a 3-fold error: 69.4%) using chemical structure alone. Our model is available at https://drumap.nibiohn.go.jp/fubrain/, and all of our data sets can be obtained from the Supporting Information.