Density functional theory and machine learning as tools to study fluorine containing molecules

Thesis

Computational organic chemistry has grown vastly in the last two decades, with the development of larger and more powerful models and more accessible experimental datasets. Advances in both Density Functional Theory (DFT) and Machine Learning (ML) have resulted in the ability to predict reaction selectivities and a range of molecular properties with ever-increasing levels of accuracy.

This thesis focuses on different DFT and ML approaches to predict enantioselectivity, NMR shifts and coupling constants for a range of fluorine-containing small molecules. In Chapter 3, both DFT and ML are used to predict the enantioselectivity of the Hydrogen Bonding Phase Transfer Catalysis (HBPTC) reaction. We then interrogate the model’s learnt parameters to suggest a selection of potential catalysts and substrates that maximise the enantioselectivity for the synthesis of alkyl beta-fluoroamines. Chapter 4 further explores the study of the HBPTC complexes by developing an approach to calculate the 19F NMR shifts of the fluoride anion and the 1JHF coupling constants across the H-F hydrogen bond. Finally, in Chapter 5, a BERT model for the prediction of 19F NMR in a range of organic molecules is introduced. Integrated Gradients (IGs) are then used to understand if the model has learnt chemistry. Finally, the model is applied to identify which regioisomer is produced from late-stage fluorination reactions purely from 19F NMR.

This thesis shows how a range of different ML and DFT options can be utilised by computational chemists to solve a series of challenging chemical problems, while also giving direction for future research in the field.

Authors: Watts, T

• DOI: 10.5287/ora-qzp8wdogp
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: computational chemistry; chemistry; machine learning; fluorine
• Subjects: Chemistry