European Research Council · Starting Grant · 2024–2029
DeepRxn — Deep learning of chemical reactions
DeepRxn is a research project funded by an ERC Starting Grant, awarded to Asst. Prof. Esther Heid in 2024. It develops new, data-driven deep-learning frameworks that predict how organic and enzymatic reactions proceed — their products, rates, and, above all, their chemo-, regio-, and stereoselectivity.
- €1.5M ERC funding
- 5 yrs 2024–2029
- 2024 Starting Grant awarded
The challenge
Exploring reactions is a central topic in chemistry, yet compared to the success of machine learning for molecules, the modeling of reactions is lagging behind — especially for stereo- and regioselective reactions. Current efforts toward sustainable synthesis, such as asymmetric organocatalysis and biocatalysis, rely on accurately predicting enantio- and regioselective reaction pathways. Meeting that need calls for new modeling approaches that go beyond the two-dimensional representations, reaction rules, and fixed three-dimensional inputs that today's methods depend on.
Our approach
DeepRxn develops new, data-driven deep-learning frameworks for modeling organic and enzymatic reactions, focusing on the chemo-, regio-, and stereoselectivity that arises through intermolecular interactions with the reagent, solvent, or catalyst. We extend molecular graph-convolutional neural networks and graph transformers to reactions in a rule-free manner, and introduce hidden three-dimensional representations that capture stereochemistry and account for atom mapping on the fly. The result is a versatile, open-source toolbox for reaction deep learning.
The DeepRxn project tackles three key areas:
- Reaction energetics Activation energies
Predicting barrier heights and transition state geometries fast enough for high-throughput screening.
- Biocatalysis Enzymatic pathways
Exploring biocatalytic routes and supporting the data-driven design of enzymatic cascades.
- Asymmetric organocatalysis Modeling stereoselectivity
Predicting enantio- and diastereoselectivity for asymmetric organocatalytic reactions.
Official project information
- Project
- DeepRxn — Deep learning of chemical reactions
- Funder
- European Research Council (ERC)
- Scheme
- ERC Starting Grant · Horizon Europe (ERC-2024-STG)
- Grant agreement
- No. 101162908
- Duration
- October 2024 – September 2029
- EU contribution
- €1,499,285
- Host institution
- TU Wien — Institute of Materials Chemistry
People on the project
-
Esther Heid Principal investigator -
Jonathan Zheng Postdoc · enzymatic reactions -
Jasper De Landsheere Ph.D. · reaction property prediction - LG Leon Ganser Ph.D. · asymmetric organocatalysis
Past contributors
-
Johannes Karwounopoulos Postdoc (2025–2026) -
Charlotte Gerhaher Ph.D. (2025–2026) -
Konstantin Mark Ph.D. (2025–2026) -
Maximilian P.-P. Kovar Ph.D. (2026)
Work supported by the ERC
Publications funded by this grant. Our full list is on the Publications page.
2026
- GRACE-OFF: A Machine-Learned Interatomic Potential for Organic Liquids Using the GRACE Architecture
- Bridging CGR representations and language models for reaction property prediction
- Toward On-the-Fly Prediction of Reaction Energetics for High-Throughput Screening
- ChemTorch: A Deep Learning Framework for Benchmarking and Developing Chemical Reaction Property Prediction Models
2025
- Feynman-Kac-Flow: Inference Steering of Conditional Flow Matching to an Energy-Tilted Posterior
- Graph-based prediction of reaction barrier heights with on-the-fly prediction of transition states
- GoFlow: efficient transition state geometry prediction with flow matching and E(3)-equivariant neural networks
Acknowledgment
Funded by the European Union (ERC-2024-STG, Project 101162908—DeepRxn). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them.