A Random Forest Regression Model for Predicting Blood–Brain Barrier Permeability

ABSTRACT The blood–brain barrier (BBB) represents a critical physiological barrier that selectively regulates molecular transport between the bloodstream and the central nervous system (CNS), creating significant challenges for CNS drug development. In this study, a random forest (RF) regression model was established to predict BBB permeability (logBB) using a large and chemically diverse dataset of 1000 compounds. Following systematic hyperparameter optimization, the final RF model ( ntree = 60, mtry = 33) demonstrated strong predictive performance across all datasets: training set (the number [ n ] of compound being 700, coefficient of determination R 2 being 0.916, root‐mean‐square [ rms ] error being 0.293), validation set ( n = 150, R 2 = 0.670, rms = 0.396), and test set ( n = 150, R 2 = 0.666, rms = 0.408). Mechanistic analysis revealed that (1) topological polar surface area (TPSA) negatively impacts transport due to reduced passive diffusion through lipid membranes, (2) moderate lipophilicity (MLOGP2) enhances permeability by balancing membrane partitioning and aqueous solubility, and (3) specific structural features—particularly halogen substitutions (CF/CCl/CBr) improve permeability through favorable hydrophobic interactions while polar/charged groups (e.g., sulfonamides, quaternary amines) hinder transport. This study not only provides a reliable and interpretable computational tool for CNS drug design but also offers quantitative structural guidelines for optimizing BBB permeability in drug development.