Predicting blood–brain barrier permeability for CNS drugs

Context

As part of an academic–industry collaboration spanning Mexico and Australia, I led a data science initiative to evaluate marine-derived kinase inhibitors (MDKIs) for their potential as CNS-active drugs. The focus was to identify kinase-targeting compounds capable of crossing the blood–brain barrier (BBB), a critical challenge in treating neurodegenerative disorders like Alzheimer’s disease.

Client/Partner Type

Academic–Industry Collaboration (Supported by UQ, Macquarie Uni and Noscira S.A.).

Challenge

Out of 471 known MDKIs, none had been systematically evaluated for BBB permeability — a major barrier for CNS drug development. Experimental assessment was unfeasible due to resource constraints and compound availability.

Solution

I designed a QSPR (quantitative structure–property relationship) modeling pipeline using open-source cheminformatics tools. The approach involved:

• Compiling a curated dataset of 471 MDKIs alongside CNS-penetrant drugs and kinase inhibitors

• Calculating over 200 molecular descriptors using RDKit

• Building and benchmarking 18 regression and classification models (e.g., logistic regression, random forest, gradient boosting)

• Optimizing top classifiers through multicollinearity filtering, feature elimination, and hyperparameter tuning

• Predicting BBB permeability with domain applicability validation using Mahalanobis distances

Outcome

• Identified 13 MDKIs with high predicted BBB permeability

• Published results in Marine Drugs :

  Plisson, F.* & Piggott, A. M. Predicting blood-brain barrier permeability of marine-derived kinase inhibitors using ensemble classifiers reveals potential hits for neurodegenerative disorders. Marine Drugs 2019, 17, 81. [DOI] [Github]

• Demonstrated the feasibility of integrating AI/ML with natural product drug discovery for CNS applications.

• Inspired the seminal and the most comprehensive review of artificial intelligence technologies (machine learning and natural language processing) applied to natural product drug discovery and molecular design.

  Saldivar-González, F. I.; Aldas-Bulos, V. D.; Medina-Franco, J. L. & Plisson, F.*. Natural product drug discovery in the artificial intelligence era. Chemical Science 2022, 13, 1526-1546. [DOI]

   

Tools

RDKit, Scikit-learn, Python, PCA, Random Forest, Gradient Boosting, QSPR Modeling, ADMET Filtering.

N.B. Established a robust, reproducible cheminformatics workflow now serving as the foundation for Ingenie Bio’s ADMET prediction services.