Gelation and aqueous solubility of small molecules with excipients: prediction using fluorimetric and computational methods
Publication Date
July 27, 2026
Creator
Maguire, Oran
Abstract
This thesis investigates experimental and computational approaches for measuring and predicting supramolecular gelation and aqueous solubility in small-molecule systems. The work combines miniaturised fluorimetric screening with scaffold-aware molecular modelling to examine how experimental design and validation strategy influence the reliability of molecular property prediction.
Chapter 2 investigates a focused library of coumarin-derived thermogelators and reference systems used to study solvent-exchange, pH-triggered, and thermally induced gelation. Miniaturised fluorimetric workflows using environment-sensitive dyes enabled high-throughput screening and revealed pathway-dependent optical responses that varied with trigger conditions and assembly history. The chapter also evaluates the limitations of semi-automated spectral analysis in heterogeneous systems, showing that scattering artefacts, overlapping signals, and evolving spectral profiles can undermine peak-fitting approaches. The findings support the interpretation of fluorescence as a probe of local molecular self-assembly rather than a direct measure of bulk mechanical gelation.
Chapter 3 investigates scaffold-aware validation strategies for low-data molecular property prediction. Scaffold abstraction and scaffold-aware splitting were used to examine how structurally related molecules should be grouped when assessing performance on unseen chemical series, providing a more realistic assessment of generalisation than random partitioning alone.
Chapter 4 investigates low-data aqueous-solubility Quantitive Structure Property Relationship (QSPR) as a proxy modelling problem. Factorial analyses of modelling workflows showed that preprocessing, validation strategy, training data, and feature choice often influenced predictive performance more strongly than the choice of machine-learning model itself. Across aqueous-solubility benchmarks, descriptor-based models provided robust performance under low-data conditions.
Together, the work highlights the importance of physically informed experimentation and realistic validation for robust molecular property prediction in diverse molecular systems.
Item Type
ethesis
Thesis Type
PhD
Supervisors
University of Nottingham
Marlow, Maria
University of Nottingham
Univerisity of Nottingham
Associated Schools / Departments
UoN Repository URI
Except where otherwise noted, this item's license is described as
File(s)![Thumbnail Image]()
Name
OranMaguire-paxom1-PhDThesisSubmission.pdf
Size
15.9 MB
Format
Adobe PDF
Checksum (MD5)
88bbf56d0fa78993dc51c6883842aa0f