Sorthouse, DavidAlharthi, Bashayr2024-12-042024Harvardhttps://hdl.handle.net/20.500.14154/74009Intestinal permeability (Peff) is a critical determinant of drug absorption and plays a pivotal role in the accuracy of Physiologically Based Pharmacokinetic (PBPK) models, which are essential for drug development and regulatory approval. Accurate prediction of Peff is vital for optimizing these models, enhancing bioavailability predictions, and reducing the reliance on extensive in vivo testing, thereby accelerating the drug development process. In this study, a machine learning (ML) model was developed to predict Peff using a small and complex dataset derived from pharmacokinetic studies. Given the dataset’s limitations, including its small size and inherent variability, traditional ML approaches were insufficient. To address these challenges, data augmentation (DA) techniques, particularly synthetic data generation, were employed to enhance the training data, thereby improving the model’s predictive performance. The suitability of the augmented data was evaluated by comparing model performance on synthetic versus real data. The study further explored the benefits and limitations of data augmentation, demonstrating its potential to address small dataset issues in pharmacokinetic modelling. Model interpretation was also conducted to understand the relationships between the target (Peff) and feature variables, offering insights into the key factors influencing the model’s predictions. This research underscores the value of data augmentation in enhancing the predictive power of ML models for PBPK applications and lays the groundwork for future work focused on optimizing data augmentation strategies and refining model interpretability.27enPharmaceuticsThesis