In Silico Prediction of Skin Permeability Using a Two-QSAR Approach

Area of Science:

• Pharmacology
• Computational Chemistry
• Biophysics

Background:

• Topical and transdermal drug delivery are crucial administration routes.
• Skin permeability is a key parameter in drug discovery and development.
• Ex vivo human skin models are vital for predicting in vivo skin permeability.

Purpose of the Study:

• To develop a predictive model for skin permeability coefficient using a novel two-Quantitative Structure-Activity Relationship (QSAR) scheme.
• To elucidate the intrinsic mechanisms governing skin permeation.
• To enhance drug discovery and development processes through accurate permeability prediction.

Main Methods:

• Employed a machine learning-based hierarchical support vector regression (HSVR) model for prediction.
• Utilized classical partial least square (PLS) for uncovering permeation mechanisms.
• Integrated HSVR and PLS using ex vivo human skin permeability data from literature.

Main Results:

• The HSVR model demonstrated superior predictive performance compared to PLS across training, test, and outlier sets.
• HSVR exhibited consistent performance, even when tested with challenging mock data.
• PLS successfully identified interpretable relationships between molecular descriptors and skin permeability.

Conclusions:

• The synergistic combination of predictive HSVR and interpretable PLS models offers a powerful tool for drug discovery.
• This integrated approach facilitates accurate prediction of skin permeability and understanding of drug permeation mechanisms.
• The developed methodology aids in optimizing topical and transdermal drug development strategies.