Two contrasting machine learning approaches to solving NMR spectra for molecular constitution and molecular configuration are presented in this doctoral thesis. The first is a unique, proof-of-principle demonstration of the direct computation of molecular structure from NMR data. It is shown how model features can be tailored to successfully map this relationship by machine learning, and where issues arise implementing the method in an experimental setting. The second approach is the collaborative advancement of the second generation of IMPRESSION – a published machine learning NMR prediction tool. Dataset expansion is carried out to improve the model’s performance and extend its applicability domain. An experimental case study is also outlined to demonstrate how IMPRESSION can be used in tandem with rapid semi-empirical geometry optimisation to solve a real world stereochemical assignment in a short time frame.
| Date of Award | 9 Dec 2025 |
|---|
| Original language | English |
|---|
| Awarding Institution | |
|---|
| Sponsors | Evotec (UK) ltd |
|---|
| Supervisor | Stephen Mann (Supervisor) & Craig P Butts (Supervisor) |
|---|
- NMR
- machine learning
- Cheminformatics
- automated structure elucidation
- Graph Neural Network
- IMPRESSION
The Solution of NMR Spectra by Machine Learning: Modelling the Forward and Inverse Relationship between NMR Properties and Chemical Structure
Honore, B. T. (Author). 9 Dec 2025
Student thesis: Doctoral Thesis › Doctor of Philosophy (PhD)