Machine Learning Exploration of the Relationship Between Drugs and the Blood–Brain Barrier: Guiding Molecular Modification

Abstract

Objective

This study aimed to improve the efficiency of pharmacotherapy for CNS diseases by optimizing the ability of drug molecules to penetrate the Blood-Brain Barrier (BBB).

Methods

We established qualitative and quantitative databases of the ADME properties of drugs and derived characteristic features of compounds with efficient BBB penetration. Using these insights, we developed four machine learning models to predict a drug's BBB permeability by assessing ADME properties and molecular topology. We then validated the models using the B3DB database. For acyclovir and ceftriaxone, we modified the Hydrogen Bond Donors and Acceptors, and evaluated the BBB permeability using the predictive model.

Results

The machine learning models performed well in predicting BBB permeability on both internal and external validation sets. Reducing the number of Hydrogen Bond Donors and Acceptors generally improves BBB permeability. Modification only enhanced BBB penetration in the case of acyclovir and not ceftriaxone.

Conclusions

The machine learning models developed can accurately predict BBB permeability, and many drug molecules are likely to have increased BBB penetration if the number of Hydrogen Bond Donors and Acceptors are reduced. These findings suggest that molecular modifications can enhance the efficacy of CNS drugs and provide practical strategies for drug design and development. This is particularly relevant for improving drug penetration of the BBB.

Graphical Abstract