Predicting Regioselectivity in Pd-Catalysed C–H Functionalisation of Aromatic Heterocycles

Abstract

Pd-catalysed C–H functionalisation reactions offer a greener alternative to traditional
cross coupling reactions however the ubiquity of the C–H bond means that
regioselectivity is difficult to control in the absence of directing groups.
One approach to overcome this is to exploit a substrate’s innate reactivity. This
approach works particularly well with N-containing heteroarenes; their unique
electronic properties differentiate C–H sites from one another which often gives high
selectivity in C–H functionalisation. Despite this, it is not trivial to identify these
innate sites before experimental validation and no predictive tool currently exists to
do this.
The work discussed in Chapter 2 shows that the regioselectivity of such substrates
depends on three properties: C–H acidity, C–H nucleophilicity and N donating
strength (donicity), all of which can be calculated using Density Functional Theory
(DFT). These metrics were calculated for a complete dataset of N-heteroarenes and
were used to select key structures for experimental validation of regioselectivity in
C–H arylation reactions. This synergy of computation and experiment allows the
derivation of a decision tree that can be used to relate the DFT metrics to
regioselectivity predictions with complete agreement to experiment.
To build a predictive tool, machine learning models were trained to predict the three
DFT metrics and classify each structure according to its regioselectivity. Kernel Ridge
Regression (KRR) and Random Forest (RF) models were found to give the highest
performance of metric prediction and classification respectively, with the
combination of these models making predictions of experimental regioselectivity
with 88% accuracy. This gives the CH Arylation: Machine Predictions of Selectivity
(CHAMPS) architecture, which is detailed in Chapter 3.
The C2 selective C–H arylation of imidazo[1,2-a]pyrimidines is investigated in Chapter
4. This reaction is thought to proceed via C3 arylation followed by Dimroth
rearrangement and is low yielding (39%). ReactIR studies suggest the formation of
degradation products that have not yet been isolated

Date of Award	23 Jan 2024
Original language	English
Awarding Institution	University of Bristol
Supervisor	Robin B Bedford (Supervisor), Craig P Butts (Supervisor) & Natalie Fey (Supervisor)