Explorations into modeling human oral bioavailability.

Area of Science:

• Pharmacokinetics
• Drug Discovery
• Computational Chemistry

Background:

• Human oral bioavailability is a critical factor in drug development.
• Accurate prediction of bioavailability aids in selecting viable drug candidates.
• Existing models using large, diverse datasets show limited predictability.

Purpose of the Study:

• To investigate if modeling human oral bioavailability improves when compounds share specific ADME properties.
• To develop enhanced quantitative models for predicting oral bioavailability.
• To assess the impact of pharmacokinetic property similarity on model performance.

Main Methods:

• Utilized a dataset of 772 drug compounds.
• Employed Kohonen's self-organizing Neural Network (KohNN) for training and test set selection.
• Applied multiple linear regression (MLR) analysis to build quantitative models.
• Developed models for the whole dataset and for four distinct subsets based on ADME properties.

Main Results:

• A model built on the entire dataset exhibited limited predictive power.
• Models developed from subsets of compounds with shared ADME properties demonstrated improved predictability.
• MLR analysis on specific subsets yielded better quantitative models for bioavailability.

Conclusions:

• Grouping drug compounds by shared pharmacokinetic properties enhances the accuracy of oral bioavailability prediction models.
• This approach offers a more effective strategy for developing reliable predictive models in drug discovery.
• Tailoring models to specific compound subsets is crucial for improving pharmacokinetic predictions.