Research
Current research
Research in the Habershon group focuses on development and application of new methods for modelling chemical dynamics in complex many-particle systems. Our interests span catalysis, protein-folding, nanoparticles, photochemistry and reaction rate evaluation.

Automated reaction discovery
In the last few years, we have developed a set of graph-driven sampling (GDS) methods to enable searches for reaction paths in complex chemical systems [See:J. Chem. Phys., 143, 094106 (2015).Link opens in a new windowand J. Chem. Theory Comput., 12, 1786 (2016)Link opens in a new window]. Our GDS strategy enables us to rapidly build up a chemical reaction network (CRN) describing the emergent reactive chemistry of complex molecular systems.
Our approach also enables determination of double-ended reaction-mechanisms, providing a route to testing mechanistic hypothesis in a direct and automated fashion [See:J. Phys. Chem. A, 123, 3407-3417 (2019) Link opens in a new windowand Cat. Sci. Tech., 9, 6357-6369 (2019)Link opens in a new window].
In 2024, we released the Kinetica.jl simulation code - this integrates automaed reaction discovery with kinetics-driven exploration and truncation of reaction networks, driven by machine-learned reaction-rate predictors.
For more details, see our Software page.
Accelerated protein folding
We have recently shown how our GDS approach - originally developed for reaction discovery - can be adapted to enable fast generation of candidate protein-folding paths.
By mapping protein-folding pathways into the reduced dimensional space spanned by the residue contact-map, we have shown how GDS enables pathways comparable to brute-force MD to be generated.We have also shown how GDS can be used to capture folding pathways in proteins with heterogeneous available folding paths.
Physics-informed program synthesis
We have recently shown how automated program synthesis - a kind of `self-writing code' similar to symbolic regression - can be used to generate entirely new algorithms that give good approximate solutions to the vibrational Schrödinger equation. Here, a "code assembler" acts to construct new candidate algorithms, and subsequently optimises their performance using comparison against exact results for model systems as a performance metric.
We have also demonstrated how this same strategy can be applied to predict anharmonic vibrational spectra for small molecular systems, and we are currently working on expanding the system-sizes that can be treated with these algorithms.
J. Chem. Theory Comput.,21, 307–320 (2024)
