Explaining Blood-Brain Barrier Permeability by Synergistic Effect on Molecular Substructures

he blood-brain barrier permeability (BBBP) is a critical factor in the design of drug candidates, particularly those designed to either target or avoid interaction with the central nervous system. However, explaining BBBP requires considering several underlying mechanisms and numerous molecular substructures that influence this property, ultimately hindering the development of a fully interpretable understanding of the BBBP process. In this study, we propose an interpretable machine learning algorithm that can explain BBBP based on synergistic effects on molecular substructures. Furthermore, we have applied our approach to mechanisms related to BBBP. Using the relative importance analysis for each substructure, we were able to identify synergistic groups that either positively or negatively affect the target property. This allowed us to screen for molecules with multiple positive or negative effects on the target property. We believe that our approach provides both interpretable and predictive models for the design of drug candidates.