Discovery of synthesisable organic materials
File(s)
Author(s)
Bennett, Steven
Type
Thesis
Abstract
Materials development is often synonymous with technological advancement, accompanying the progression of humanity, and can help address many contemporary issues humans face, including climate change, pollution, and resource scarcity. With their unique structural properties and vast surface areas, porous materials have emerged as promising materials for addressing some of these challenges, with applications in gas separation, storage, catalysis, and sensing. Porous organic cages are a sub-class of porous materials that have gathered significant attention in recent years due to their advantages over porous framework materials and the diverse range of geometric shapes that cages can form. However, the discovery of porous organic cages often relies on incremental modifications to known precursors or serendipitous discoveries. Computational screening can assist but is often hindered by the inability to experimentally realise many of the computationally predicted candidates.
In this thesis, I present a computational workflow to bridge the gap between computational prediction and synthetic realisation of porous organic cages. I developed a machine learning model to mimic the decision-making process of expert chemists when selecting synthesisable cage precursors. I used this model to filter a large library of cage precursors and used computational simulations to determine whether each pair of precursors would form a cage with a permanent internal cavity.
Next, I used the predictions from this computational screening workflow to guide the creation of a precursor library of diamines and trialdehydes, which were combined using high-throughput experimentation. To analyse the outcome of the 144 reactions performed, I developed an automated reaction analysis workflow to determine whether a cage formed. By combining this workflow with computational simulations, I was able to predict which precursor combinations would form shape persistent cages, identifying candidates that were expected to be porous. The automated reaction analysis workflow in my thesis provides a foundation to expand the number of reactions that can be performed and analysed in parallel using high-throughput experimentation, which could, in time, lead to the prediction of cage assembly using data-driven techniques.
In this thesis, I present a computational workflow to bridge the gap between computational prediction and synthetic realisation of porous organic cages. I developed a machine learning model to mimic the decision-making process of expert chemists when selecting synthesisable cage precursors. I used this model to filter a large library of cage precursors and used computational simulations to determine whether each pair of precursors would form a cage with a permanent internal cavity.
Next, I used the predictions from this computational screening workflow to guide the creation of a precursor library of diamines and trialdehydes, which were combined using high-throughput experimentation. To analyse the outcome of the 144 reactions performed, I developed an automated reaction analysis workflow to determine whether a cage formed. By combining this workflow with computational simulations, I was able to predict which precursor combinations would form shape persistent cages, identifying candidates that were expected to be porous. The automated reaction analysis workflow in my thesis provides a foundation to expand the number of reactions that can be performed and analysed in parallel using high-throughput experimentation, which could, in time, lead to the prediction of cage assembly using data-driven techniques.
Version
Open Access
Date Issued
2023-05-24
Date Awarded
01/08/2023
License URL
Advisor
Jelfs, Kim
Greenaway, Rebecca
Sponsor
The Leverhulme Trust
Publisher Department
Chemistry
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)