About_

I'm a DPhil (PhD) student in the Department of Statistics, broadly interested in machine learning approaches to biomolecular structure and design. My research in the Oxford Protein Informatics Group (OPIG) focusses on advancing de novo antibody design using generative models. In this work, I am fortunate to be supervised by Prof Charlotte Deane, and Matthew Raybould, as well as Rebecca Croasdale-Wood at AstraZeneca.

In the past, I've tackled enzyme-substrate interaction prediction using protein language models, and also had a brief stint in theoretical chemical physics, predicting quantum tunnelling splittings using classical molecular dynamics. Check out the research and publications pages for more details.

As a hobby, I've occasionally dabbled in information-theoretic approaches to evolution, and game theory on graphs.

My DPhil work is generously funded by Oxford University's flagship Clarendon Scholarship, alongside additional partnership awards by Oxford University, Balliol College, and AstraZeneca. Throughout my BSc and MSc, I had the privilege to receive support from the German Academic Scholarship Foundation.



Education_

University of Oxford logo

2023-2027

DPhil at the Department of Statistics (SABS:R3 CDT)

Balliol College, University of Oxford


logos of ETH Zurich, University of Zurich, and University of Basel

2020-2023

MSc Computational Biology & Bioinformatics

ETH Zürich, Switzerland


Heidelberg University logo

2015-2018

BSc Biochemistry

Heidelberg University, Germany

Past Research Projects_

Schematic of Model Architecture

Enzymatic catalysis is a cornerstone of biological systems, but the secondary substrates of most enzymes are unknown.

In my MSc thesis project, I developed a recommendation system based on neural matrix factorisation that scores potentially interacting enzyme-substrate pairs across all of enzyme and compound space. To do so, it uses only the substrate's MACCS fingerprint and a language-model embedding of the enzyme's sequence.

The model can be used to identify likely-interacting pairs in order to accelerate high-value applications like drug-lead and -target identification, or drug re-purposing.

This work was carried out in the Learning and Adaptive Systems Group at the AI Center of ETH Zürich, under the supervision of Prof Andreas Krause, Jonas Rothfuss, and Mojmir Mutný.


Hydronium PIMD

Tunnelling effects can strongly influence chemical reaction rates and tunnelling splittings give rise to characteristic features in high-resolution molecular spectra. Unfortunately, quantum calculations often scale unfavourably with system size.

To circumvent this, I helped develop and implement a new method to calculate small-molecule tunnelling splittings without quantum mechanics, using path-integral molecular dynamics and free-energy methods adopted from biomolecular simulation.

This work was carried out in the Theretical Molecular Quantum Dynamics Group at ETH Zürich, under the supervision of Prof Jeremy Richardson and Dr George Trenins. Read all about our work in the associated publication.



Spectrin Repeat At Point of Rupture

Spectrins are a family of filamentous proteins composed of repeating, triple-helical repeat domains. They act as cytoskeletal springs, allowing cells to recover their shape after elastic deformation.

In my BSc thesis project, I used molecular dynamics simulations to study the mechanical properties of these repeating domains in isolation. In particular, I focussed on their unfolding and re-folding behaviour under mechanical tension.

This work was carried out in the Molecular Biomechanics Group at the Heidelberg Institute for Theoretical Studies (HITS), under the supervision of Prof Frauke Gräter and Dr Csaba Daday.

Publications_

You can also find me on Google Scholar and ORCID.

Challenges and compromises: Predicting unbound antibody structures with deep learning.
A. Greenshields-Watson; O. Vavourakis; F. C. Spoendlin; M. Cagiada; C. M. Deane
Curr. Opin. Struct. Biol.; February 2025; 90: 102983. doi: 10.1016/j.sbi.2025.102983


Exact tunneling splittings from symmetrized path integrals.
G. Trenins, L. Meuser, H. Bertschi, O. Vavourakis, R. Flütsch, J. O. Richardson
J. Chem. Phys.; 21 July 2023; 159 (3): 034108. doi: 10.1063/5.0158879