A novel in silico approach for predicting unbound brain-to-plasma ratio using machine learning-based support vector...

Area of Science:

• Pharmacology
• Computational Chemistry
• Neuroscience

Background:

• The blood-brain barrier (BBB) is a critical biological interface controlling substance entry into the central nervous system (CNS).
• Accurate prediction of BBB penetration is vital for both CNS-targeted therapeutics and systemic drugs to manage side effects.
• Existing methods for assessing BBB penetration can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop and validate a novel in silico model for predicting unbound brain-to-plasma concentration ratio (K_p,uu,brain).
• To utilize a machine learning approach, specifically hierarchical support vector regression (HSVR), for enhanced prediction accuracy.
• To provide a reliable tool for early-stage drug discovery, streamlining the assessment of BBB penetration.

Main Methods:

• Development of a hierarchical support vector regression (HSVR) model.
• Training and validation using a diverse dataset of compounds with known BBB penetration properties.
• Rigorous validation using comprehensive metrics and a mock test for practical application assessment.

Main Results:

• The HSVR model demonstrated high accuracy, predictivity, and robustness in predicting K_p,uu,brain values.
• Comparative analysis showed the HSVR model significantly outperforms existing published models.
• The model effectively captures complex BBB penetration mechanisms involving diffusion and active transport.

Conclusions:

• The developed HSVR model offers a reliable and efficient computational tool for predicting BBB penetration.
• This approach can accelerate drug discovery by minimizing reliance on in vivo studies and supporting the "fail early, fail fast" paradigm.
• The model facilitates the identification of promising CNS drug candidates and helps mitigate CNS-related challenges for systemic drugs.