Prof. Dr. Fernanda Duarte, Department of Chemistry, Oxford, UK

"Reaction modelling and synthesis planning: from mechanistic insights to data-driven exploration"

Abstract:

Predicting the outcomes of chemical reactions remains a fundamental challenge in organic and computational chemistry. Traditional quantum‑chemical methods have become well-established tools for rationalising mechanisms and guiding experiments, yet they remain computationally intensive and often limited to expert users. As chemical systems grow in complexity and require the exploration of multiple competing pathways, these limitations become increasingly prohibitive.

In this seminar, I will discuss how combining automated mechanistic exploration with modern machine learning tools offers new possibilities for understanding and predicting organic reactivity. I will highlight our group’s development of autodE, an open‑source framework that automates reaction mechanism exploration. [1,2] I will then describe our efforts to model solution-phase reactivity using machine learning interatomic potentials (MLPs) as efficient surrogates for traditional ab initio methods [3], enabling the exploration of complex transformations, including organometallic processes relevant to catalysis.

Where full mechanistic exploration is impractical, data-driven approaches offer a complementary route. I will showcase our work on reactivity prediction and heterocycle synthesis planning, demonstrating how curated reaction datasets and fine-tuning strategies can improve model performance. [4,5] I will conclude by reflecting on the evolving interplay between mechanistic insight and data-driven methodologies as the field continues to advance.

References:

1.   T. A. Young, J. J. Silcock, A. J. Sterling, F. Duarte. autodE: Automated Calculation of Reaction Energy Profiles— Application to Organic and Organometallic Reactions. Angew. Chem. Int. Ed. 2021, 60, 4266

2.   S. R. Maiti, David Buttar, F. Duarte. Benchmark of double-ended transition state search methods for metal-catalysed reactions. ChemRxiv 2025

3.  (a) H. Zhang, V. Juraskova, F. Duarte. Modeling Chemical Processes in Explicit Solvents with Machine Learning Potentials. Nat. Commun.,2024, 15, 6114. (b) V. Vitartas, H. Zhang, V. Juraskova, T. Johnston-Wood, F. Duarte. Active learning meets metadynamics: Automated workflow for reactive machine learning potentials Digit. Discovery 2026, 5, 108

4.   (a) E. Wieczorek, J. W. Sin, M. T. O. Holland, L. Wilbraham, V. S. Perez, A. Bradley, D. Miketa, P. E. Brennan, F. Duarte. Transfer learning for Heterocycle Synthesis Prediction. J. Chem. Inf. Model. 2025, 65, 15, 7851; (b) S. Tanovic, E. Wieczorek, F. Duarte. An exploration of dataset bias in single-step retrosynthesis prediction Dig Discovery 2026

Biography:

Fernanda Duarte is an Associate Professor in the Department of Chemistry at Oxford. She obtained her PhD from the Pontificia Universidad Católica de Chile in 2012. As a postdoctoral researcher, she trained in biomolecular modelling with Prof. Lynn Kamerlin at Uppsala University and in computational organic chemistry with Prof. Robert Paton at the University of Oxford as a Newton Fellow. After a brief period as a lecturer at the University of Edinburgh, she returned to Oxford in 2018 as an Associate Professor. Her research combines software development (e.g. autodE, Cgbind and mlp-train) with practical applications, from enzymes to small molecules, focusing on mechanistic modelling and molecular design.  Fernanda has received several awards, including the MGMS Frank Blaney Award (2020), the OpenEye Outstanding Junior Faculty Award (2021), the Harrison-Meldola Memorial Prize (2021), and the Novartis Early Career Award in Chemistry (2022). She was also selected as a Finalist for Chemical Sciences in the 2024 Blavatnik Award in the UK.