An integrated computational methodology with data-driven machine learning, molecular modeling and PBPK modeling to...

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Drug Delivery

Background:

• Solid dispersions (SD) enhance drug bioavailability but their development is inefficient, relying on random screening.
• Current methods lack predictive power for formulation performance and in vivo outcomes.

Purpose of the Study:

• To develop and validate an integrated computational methodology for accelerating solid dispersion formulation development.
• To predict dissolution profiles, molecular interactions, and human pharmacokinetic behavior of SD formulations.
• To establish an in silico approach that surpasses traditional trial-and-error laboratory methods.

Main Methods:

• Machine learning (ML) models (random forest) were trained on 674 SD dissolution profiles to predict dissolution behavior.
• Molecular dynamics (MD) simulations were used to investigate polymer-drug interactions at the molecular level.
• Physiologically based pharmacokinetic (PBPK) modeling was employed to predict human pharmacokinetic profiles.

Main Results:

• ML models demonstrated good prediction performance for classifying dissolution profiles and predicting dissolution times.
• MD simulations revealed that HPMCAS-based SD formulations exhibited faster dissolution than Eudragit-based ones, aligning with experimental data.
• PBPK modeling successfully predicted the in vivo pharmacokinetic profile of a vemurafenib-HPMCAS SD formulation.

Conclusions:

• An integrated computational approach combining ML, MD, and PBPK modeling effectively predicts the in vitro and in vivo performance of solid dispersion formulations.
• This in silico methodology significantly enhances the efficiency and predictability of pharmaceutical formulation development.
• The developed tools facilitate prediction of formulation composition, in vitro release, and in vivo absorption, streamlining drug development.