Advancing data-driven and mechanistic modelling for oral drug product development

Simple item page
dc.check.chapterOfThesisPlease redact the following for two years: Chapter 4 Chapter 6 Chapter 7 Please do not make the .zip Latex source, "MurrayJD_PhD2026_Latex.zip," available for public download.en
dc.check.date2028-12-31
dc.contributor.advisorGriffin, Brendan T.
dc.contributor.advisorO'Dwyer, Patrick
dc.contributor.advisorexternalBennett-Lenane, Harriet
dc.contributor.authorMurray, Jack D.en
dc.contributor.funderResearch Irelanden
dc.contributor.funderFulbright Commission in Irelanden
dc.date.accessioned2026-09-29T15:18:07Z
dc.date.available2026-09-29T15:18:07Z
dc.date.issued2026-05-12
dc.date.submitted2026-05-12
dc.description.abstractDespite the widespread application of modelling to accelerate and de-risk oral drug product development, the modelling approaches themselves have not kept pace with the demands of the field. In particular, the paucity of data, the high-dimensional product design space, and the noise in pharmacokinetic and experimental measurements limit the effectiveness of the prevailing modelling approaches. This thesis responds to these challenges by developing novel predictive strategies tailored to the realities of drug product development, with an emphasis on improving accuracy, interpretability, and mechanistic understanding. Chapter 2 critically analysed machine learning models reported in the pharmaceutics literature, with a view to developing best practice guidelines for data-driven modelling. Recurrent limitations were identified, including inadequate sharing of code, a preference for "black-box'' models, and insufficient mining of the drug development datasphere. Consequently, Chapter 3 addressed the lack of a unified drug product database. Regulatory filings were retrieved programmatically from the European Medicines Agency and then processed to extract formulation, pharmacokinetic, and clinical information. This was further linked to external databases to ensure interoperability by design. The chemical space of centrally authorised oral drugs showed poor agreement with established drug-likeness criteria, while association rule learning uncovered excipient selection patterns for oral tablets. An oral fraction absorbed model built on this data achieved a test-set balanced accuracy of 0.725, despite unreliable reporting of this parameter. This chapter thus demonstrated the benefits and caveats of using highly-linked drug product data for modelling. Chapter 4 questioned both the need for complex models and the existence of only one structure-property narrative through the application of optimal sparse decision tree enumeration. By implementing the SORTeD algorithm on the MeluXina supercomputer, sets of near-optimal decision trees, termed "Rashomon sets,'' were generated for six oral developability endpoints. For oral bioavailability, 9862 trees were discovered with accuracies exceeding 0.827 on the full dataset, revealing structure-property relationships which extend beyond the existing filters of drug-likeness. Additionally, the Rashomon set provided a more robust view of feature importance when compared to the one-model paradigm by identifying descriptors that were selected consistently across many performant trees. The thesis transitions from machine learning to mechanistic modelling of membrane permeation in Chapter 5. Here, the reduced-resistances model was proposed to explain how a solubilising receiver medium can increase drug flux in hollow fibre membranes (HFM). In experiments with nine solubilising additives, flux enhancement followed the rank-order, but not the magnitude, of the receiver solubilisation capacity. At high receiver-to-donor solubility ratios, a hyperbolic relationship was observed where diffusion through the donor-side concentration boundary layer (CBL) was rate-limiting. These findings provide a rational basis for receiver selection, thereby supporting more informative assessment of formulation effects on in vitro absorption. Chapter 6 extended the mechanistic framework of Chapter 5 by considering the impact of cylindrical geometry on mass transfer in HFM. By applying cylindrical calculations to the donor-side CBL and membrane, CBL thicknesses which are physically consistent with the apparatus dimensions were obtained. The preferred model suggests that steep concentration gradients form in the donor-side CBL due to radial dilution, which has implications for particles that undergo dissolution in this region. This chapter additionally introduced a miniaturised HFM setup using the Pion MicroDISS Profiler, reducing material requirements and enabling inline drug quantification. This thesis demonstrates how integrated data-driven and mechanistic modelling strategies can accelerate oral drug product development across lead optimisation, excipient selection, and permeability assessment. The studies presented show that innovative modelling is valuable not only for prediction, but also for strengthening understanding. Ultimately, these contributions advance a mechanistically informed and data-enabled pathway for drug product development.en
dc.description.statusNot peer revieweden
dc.description.versionAccepted Versionen
dc.format.mimetypeapplication/pdfen
dc.identifier.citationMurray, J. D. 2026. Advancing data-driven and mechanistic modelling for oral drug product development. PhD Thesis, University College Cork.
dc.identifier.endpage248
dc.identifier.urihttps://hdl.handle.net/10468/19361
dc.language.isoenen
dc.publisherUniversity College Corken
dc.relation.projectResearch Ireland (GOIPG/2022/1580)
dc.rights© 2026, Jack Denis Murray.
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/
dc.subjectDrug product development
dc.subjectOral drug absorption
dc.subjectCheminformatics
dc.subjectPermeability
dc.subjectMachine learning
dc.subjectBiopharmaceuticS
dc.subjectPharmaceutics
dc.titleAdvancing data-driven and mechanistic modelling for oral drug product development
dc.typeDoctoral thesisen
dc.type.qualificationlevelDoctoralen
dc.type.qualificationnamePhD - Doctor of Philosophyen
Files
Original bundle
Now showing 1 - 2 of 2
Thumbnail Image
Name:
MurrayJD_PhD2026.pdf
Size:
4.83 MB
Format:
Adobe Portable Document Format
Description:
Full Text E-thesis
Thumbnail Image
Name:
MurrayJD_PhD2026_Latex.zip
Size:
3.91 MB
Format:
http://www.iana.org/assignments/media-types/application/zip
Description:
Latex File
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
5.2 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Research Theses
College of Medicine and Health - Doctoral Theses
Pharmacy - Doctoral Theses