Physicochemical property profile for brain permeability: comparative study by different approaches.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Pharmacology

Background:

• Accurate prediction of central nervous system (CNS) or non-CNS penetration is vital for drug development.
• Traditional medicinal chemistry approaches often yield suboptimal classification models.
• Machine learning offers alternative strategies for predicting drug brain penetration.

Purpose of the Study:

• To compare the efficacy of various classification models for predicting drug brain penetration.
• To evaluate both traditional medicinal chemistry methods and machine learning techniques.
• To identify the most accurate and interpretable models for CNS penetration prediction.

Main Methods:

• Applied ten different approaches, including seven medicinal chemistry methods (e.g., "rule of 5") and three machine learning techniques: logistic regression (LR), random forest (RF), and support vector machine (SVM).
• Utilized a training set of 1000 chemicals/drugs and an external test set of 100 drugs.
• Employed 41 diverse medicinal chemistry descriptors reflecting physicochemical properties.

Main Results:

• Medicinal chemistry approaches demonstrated poor classification accuracy and unbalanced models.
• Random forest (RF) and support vector machine (SVM) achieved 82% and 84% accuracy on the external test set, respectively.
• Logistic regression (LR) provided accuracy equivalent to RF and SVM, with added benefits of simplicity and mechanistic interpretability.

Conclusions:

• Machine learning models, particularly LR, RF, and SVM, significantly outperform traditional medicinal chemistry approaches for predicting drug brain penetration.
• Logistic regression is highly recommended for medicinal chemists due to its strong performance, simplicity, and clear mechanistic insights.
• Accurate CNS penetration prediction using LR can streamline drug discovery and development processes.