Back to search

National and Kapodistrian University of Athens

Development of novel methods in population pharmacokinetics using machine learning and fractional differential equations

Abstract

dc:description

Introduction The purpose of this dissertation is to study novel methods regarding modeling in the field of pharmacometrics and more specifically population pharmacokinetics(popPK). Modeling is a key part of pharmacokinetics and especially in describing the four key processes that occur after a drug enters a subject’s body (absorption, distribution, metabolism and excretion), personalized treatment, study design, bioequivalence and more. In order to maximize the utilization of modeling, the use of tools from other fields such as statistics, machine learning, mathematical biology etc. is required. These tools assist in drastically expanding the class of problems modeling can solve in the field of popPK. In the work presented in this thesis, the focus was on employing machine learning and fractional calculus in this discipline. The first part (Chapter 1 and Chapter 2) are focused in applying machine learning methodology in classic NLME modeling as a post-step process tackling the issue of model misspecification. The second part is a study of numerical methods for fractional differential equations and an implementation of such methods to the widely used NONMEM software that performs non-linear mixed effects modeling in the context of pharmacometrics. Chapter 1We address the problem of model misspecification in population pharmacokinetics(PopPK), by modeling residual unexplained variability (RUV) by machine learn-ing (ML) methods in a post-processing step after conventional model building. The practical purpose of the method is the generation of realistic virtual patient profiles. The proposed methodology consists of the following steps: After developing a PopPK model, the individual residual errors IRES = DV-IPRED, are computed, where DV are the observations and IPRED the individual predictions and are modeled by ML to obtain IRESML. Correction of the IPREDs can then be carried out as DVML = IPRED + IRESML. The methodology was tested in a6PK study of ropinirole, for which a PopPK model was developed while a second deliberately misspecified model was also considered. Various supervised ML algorithms were tested, among which Random Forest gave the best results. The ML model was able to correct individual predictions as inspected in diagnostic plots and most importantly it simulated realistic profiles that resembled the real data and canceled out the artifacts introduced by the elevated RUV, even in the case of the heavily misspecified model. Chapter 2Expanding on the concepts introduced in the previous chapter, we introduce a metric with the purpose of quantifying the amount of model misspecification after an initial popPK modeling step. The methodology aims to provide a numerical component in the modeler’s toolkit that informs on how well the model structure is able to describe the data that is also appropriately scaled (0 to 1) which helps interpret the results and make model comparison easier. The proposed metrics use the ML modeled residuals presented in the previous chapter, IRESML to calculate how well the ML model fitted the initial IREStot. The rationale is simple: a good fit and high metric, means that the ML model has found signal in the residuals which indicates a biased model and high misspecification. A poor and small met-ric indicate that ML has failed to fit the residuals and the initial popPK model is unbiased. The metrics -indicated R2ML and RR- were based on the classic R2and proportion of the fitted residual to the total. First, the metrics were tested on virtual case studies where a 2CMT model was used to generate synthetic data which were then fit using a deliberately misspecified 1CMT and the correct 2CMTmodel. Metrics were calculated using IRES and IWRES to check the effect the type of residual had on the final result. After validating on controlled scenarios, the metrics were then tested on several real case studies. Previously developed models were tested in this case to avoid bias. The real scenarios involved datafrom busulfan, myofetil, octreotide and the ropinirole data used in the previous chapter. Metric values were compared with typical goodness-of-fit measures and7visuals in order to validate their performance. In the extreme case studies of the synthetic data (high and minimal misspecification), they performed as expected taking high and low values respectively. For the real case studies values were also in agreement with the visual means of model assessment and the conclusions drawn from the initial modelers. Chapter 3 Fractional differential equations (FDEs), i.e. differential equations with derivatives of non-integer order, can describe certain experimental datasets more accurately than classic models and have found application in pharmacokinetics (PKs),but wider applicability has been hindered by the lack of appropriate software. In the present work an extension of NONMEM software is introduced, as a FOR-TRAN subroutine, that allows the definition of nonlinear mixed effects (NLME)models with FDEs. The new subroutine can handle arbitrary user defined linear and nonlinear models with multiple equations, and multiple doses and can be integrated in NONMEM workflows seamlessly, working well with third party packages. The performance of the subroutine in parameter estimation exercises, with simple linear and nonlinear (Michaelis-Menten) fractional PK models has been evaluated by simulations and an application to a real clinical dataset of diazepam is presented. In the simulation study, model parameters were estimated for each of 100 simulated datasets for the two models. The relative mean bias(RMB) and relative root mean square error (RRMSE) were calculated in order to assess the bias and precision of the methodology. In all cases both RMB and RRMSE were below 20% showing high accuracy and precision for the estimates. For the diazepam application the fractional model that best described the drug kinetics was a one-compartment linear model which had similar performance, ac-cording to diagnostic plots and Visual Predictive Check, to a three-compartment classic model, but including four less parameters than the latter. This is a first attempt to use FDE systems in an NLME framework, so the approach could be of interest to other disciplines apart from PKs.

Degree

thesis:*
Grantor dc:publisher
National and Kapodistrian University of Athens
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kaikousidis, Christos
  • Καϊκουσίδης, Χρήστος

Subjects

dc:subject × 10

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/58882

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Kaikousidis, Christos; Καϊκουσίδης, Χρήστος. Development of novel methods in population pharmacokinetics using machine learning and fractional differential equations. National and Kapodistrian University of Athens, 2024. http://hdl.handle.net/10442/hedi/58882