Research Interests
Research Summary
We are interested in employing electronic structure theory calculations, mainly relying on density funnctional theory (DFT), to perform computational mechanistic investigations of complex organic and organometallic catalytic systems of both academic and industrial significance. We currently focus on transition-metal catalysed C–H functionalisation as well as organocatalytic asymmetric synthesis for complex molecules constructions.We are building up capabililies in applying machine learning interatomic potentials (MLIPs) to study traditionally challenging chemical systems such as dynamic and entropic effects that are hard to capture accurately with static DFT. Our interests also include coding and workflow development to facilitate high-throughput quantum chemical studies and machine learning applications, with the goal of streamlining computational research practices.
- Computational Homogenous Catalysis
- Machine Learning for Chemistry
Organometallic catalysis
C–H functionalisation
Transition metal (TM) catalysed C–H functionalisations present many
opportunities for organic synthesis by forging useful C–X bonds from
unreactive, ubiquitous C–H bonds present in many organic molecules.
The use of TMs to selectively functionalise C–H bonds is
particularly attractive as it provides atom economical ways to either
convert small alkanes to higher valued, functionalised molecules or
directly manipulating complex molecules with other functional groups
present. Such TM-catalysed C–H functionalisation usually rely on
directing groups (DGs) for target C–H activation. Directing group free
methods have also been developed to sidestep the need for the
introduction and subsequent removal of covalent DGs for step economy.
We are interested in studying the mechanisms of such DG-assisted and
DG-free catalytic systems.
a) Transition metal-catalysed selective C–H activation using
covalent/non-covalent directing groups or innate functional
group present in substrate. b) Enantioselective C–H
functionalisation with simultaneous chirality control.
Gibbs energy profile for Pd(OAc)2-catalysed
alkynylation. MPAA ligand lowers the barrier of C–H activaiton
(ts-1' over ts-1). In addition, silver acetate
assists in lowering the barrier of β-bromide elimination
(ts-4' over ts-4).
In the study of reaction mechanisms, we often need to understand the
molecular origins giving rise to different chemical selectivities. We
may also need to compare chemical reactivities of different
substrates. For example, in the meta-selective C–H
functionalisation of arenes (top panel), we are able to
computationally elucidate the origins behind the exclusive
meta- over para- or ortho-selectivity. We do this
by comparing the turnover-frequency determining transition states
(TDTSs) at each site (middle panel). We are able to computationally
calculate the differences of the ring strain in the different sized
palladacylce intermediate using isodesmic reaction enthalpy
calculations to compare the different degrees of ring strain in each
transition state.
Highly meta-selective allylation of arenes (first panel)
and the relative barriers for arene site selectivity (middle
panel) and the ring strains for E-allyl product formation
over Z-allyl and styrenyl product formations (last
panel).
Organocatalysis catalysis
N-heterocyclic carbene (NHC) catalysis
NHC organocatalysts offer diverse catalytic activation modes: NHC-activated species can react with either electrophile at carbonyl carbon via umpolung addition or with nucleophile via oxidised Brewslow intermediate. This provides many avenues for computational tools to study the detailed mechanisms. Our lab will study metal-free organocatalysis including carbene catalysis. We aim to contribute to the field of organocatalysis for chiral molecules construction with the aim of discovering novel activation modes and functionalising challenging inert molecules such as carbonyl compounds, ketenes and alcohols for applications in pharmaceutical, agrochemical and specialty chemicals syntheses. Some challenges in the field such as using NHCs as asymmetric Brønsted bases will be explored.
c) Ambiphilic reactivity of organocarbene catalysts. d) Examples of carbene catalysed reactions using various substrates.
Chemical biology
We are also able to computationally model chemical reactions within a
biological environment. In one project, we applied DFT studies to
understand the labelling of α-C of amino acid residues, which
can be useful as potential probes for investigating chemical
structures and mechanisms. This is especially important if intact
small peptides and proteins can be labelled in situ.
Dehydroalanine (Dha) amino acid residue, formed from postsynthetic
modifications and normal metabolism, can serve as a target for
deteuration at its α-C. The mechanism for the formation of
dehydroalanine from dialkylated cysteine residue can be investigated
using quantum mechanical tools such as DFT. The competing mechanisms
of E2 elimination vs deprotonation forming a sulfonium ylide followed
by intramolecular ‘carba-Swern’ type cycloreversion can be directly
compared.
Gibbs energy profile for the competing mechanisms of Dha
formation from dialkylated cysteine residue. HOMO (at isovalue
of 0.05 a.u.) shows concerted electron movement as the formal
[3+2] cycloreversion occurs.
Traditional DFT studies in catalytic systems often rely on static DFT with implicit solvation models1. While this approach provides explanatory and predictive successes at reduced computational costs, it may fall short when solvent molecules actively participate in the reaction and explicit solvation is required. However, studying catalytic systems with explicit solvation at the DFT level can be computationally prohibitive, especially for large dynamical systems where numerous trajectories are needed.
Machine learning-accelerated molecular dynamics (MD) simulations leveraging on active learning strategy. Figure adapted and modified from Chem. Sci., 2023, 14, 8338
Building upon our work on MLIPs for reactive chemistry in heterogenous catalysis, we are currently exploring how to apply such MLIPs for studying homogenous catalytic systems where explicit solvents and large entropic changes play important roles. We hope to leverage active learning to reduce training costs without worrying about the transferability of the MLIP potential. With these tools, the study of costly and challenging systems, such as explicit solvation and dynamical effects on a large length- and time-scale, can be routinely carried out at DFT-level accuracy within affordable costs.
Research and Work Experiences
Teaching Experiences
