Identifying Substructures That Facilitate Compounds to Penetrate the Blood-Brain Barrier via Passive Transport Using...

Area of Science:

• Computational Chemistry
• Pharmacology
• Medicinal Chemistry

Background:

• The blood-brain barrier (BBB) presents a significant challenge in drug delivery, limiting the efficacy of many therapeutics.
• Understanding molecular properties that govern BBB penetration is crucial for developing effective central nervous system (CNS) drugs.

Purpose of the Study:

• To apply the Local Interpretable Model-agnostic Explanation (LIME) technique to interpret machine learning models predicting blood-brain barrier penetration.
• To identify key molecular substructures responsible for compound permeability across the BBB.

Main Methods:

• Utilized classification models including Random Forest, ExtraTrees, and Deep Residual Network trained on a BBB penetration dataset.
• Employed LIME to generate explanations for model predictions, highlighting influential molecular features.
• Filtered LIME explanations with a weight threshold of 0.1 to focus on the most relevant substructures.

Main Results:

• LIME successfully identified critical molecular substructures affecting BBB penetration.
• Nitrogenous substructures were generally found to enhance the likelihood of compounds crossing the BBB.
• The interpretability of LIME outputs demonstrated its utility in understanding BBB permeability based on molecular characteristics.

Conclusions:

• LIME provides an effective method for interpreting complex machine learning models in the context of BBB penetration.
• The identified molecular features, particularly nitrogenous groups, can guide the rational design and synthesis of novel CNS-active drugs.
• This approach offers valuable insights for the pharmaceutical industry and drug discovery research groups aiming to optimize drug candidates for BBB permeability.