Logo image
Advancing pharmacokinetic modelling : a dual approach exploring fractional order systems and neural network-based parameter prediction
Dissertation   Open access

Advancing pharmacokinetic modelling : a dual approach exploring fractional order systems and neural network-based parameter prediction

Sinenhlanhla Mtshali
Doctor of Philosophy (PHD), University of Johannesburg
2025
Handle:
https://hdl.handle.net/10210/519042

Abstract

This study combines advanced pharmacokinetic modelling techniques and machine learning-based approaches to optimise drug dosage and enhance therapeutic outcomes in clinical practice. The first component uses fractional-order differential equations to model pharmacokinetic processes, offering a novel alternative to conventional ordinary differential equations. Fractional-order models capture the complex dynamics of drug absorption, distribution, metabolism, and excretion (ADME) more accurately, particularly for drugs with non-linear kinetics or prolonged distribution phases. Two numerical methods, the Grunwald–Letnikov and L1 approximations, were validated for their ability to solve fractional models, demonstrating robust performance without analytic solutions. These methods were applied to two-compartment models for intravenous and oral drug administration routes, with sensitivity analyses revealing the significant role of fractional parameters in representing memory effects in drug kinetics. Clinical validation further supported the hypothesis that fractional models can better describe pharmacokinetics than their traditional counterparts. The second component of this research introduces a data-driven machine-learning framework for the individualised prediction of pharmacokinetic parameters. Using neural networks to address the inverse parameter estimation problem, the method integrates patient-specific data, such as medical history and physiological variables, to refine model parameters. These parameters are subsequently used in forward modelling to predict drug concentration-time profiles under various dosing regimens. The framework combines the flexibility of machine learning with the rigour of differential equation-based pharmacokinetics, achieving high accuracy in parameter estimation and close alignment between predicted and observed drug concentration profiles. Across multiple dose levels, the approach consistently demonstrated minimal error and a strong correlation between observed and predicted values. By integrating fractional modelling and machine learning, this study presents a hybrid methodology that bridges theoretical advancements in pharmacokinetics with practical, patient-centred applications. This dual approach improves the precision of drug dosage regimens and lays the groundwork for personalised medicine, enabling clinicians to optimise therapeutic strategies based on individual patient profiles.
pdf
Sinenhlanhla_Mtshali3.29 MBDownloadView
Open Access

Metrics

1 Record Views

Details

Logo image