We are thrilled to announce the winners of the Outstanding Early Career Researcher Award 2025: Daniel Probst and Jan Weinreich, recognized for their innovative paper published in Digital Discovery, “Learning on compressed molecular representations“
In their work, Daniel and Jan present MolZip, a simple yet powerful method that leverages string compression to predict molecular properties. By compressing concatenated SMILES and protein sequences, MolZip competes impressively with more complex graph-based methods outperforming 9 out of 10 state-of-the-art graph neural networks in predicting protein-ligand binding affinities.
Their approach is a compelling reminder that elegant, data-efficient solutions can still hold their ground even in the era of sophisticated machine learning models.
About the winners:
Daniel Probst received his PhD in Chemistry and Molecular Sciences from the University of Bern. After research roles at IBM and EPFL, he is now a tenure-track assistant professor at Wageningen University, focusing on sustainable machine learning applications in biology and chemistry.
Jan Weinreich specializes in machine learning for chemistry and materials science. He develops scalable algorithms for molecular property prediction and is co-founder of Chembricks AI, a company building AI platforms for materials discovery.
“Winning this prize is really cool and was a great surprise, especially for an article in one of my favourite journals. Of course, it is, as it is always the case, the contributions of scientific collectives rather than single groups or individuals that make a paper happen.”
– Daniel Probst
“We are truly honoured to receive the Outstanding Early Career Researcher Award from Digital Discovery. This recognition means a great deal to us. We hope that our work on compressed molecular representations will be of some use to the community to build AI models in chemistry that do not require huge computational resources but give quite competitive accuracy. Even with powerful GPUs at hand – sometimes being able to quickly build ML models can be helpful to “debug” data pipelines and test if allocating more attention/compute to a more “sophisticated” AI architecture”
– Jan Weinreich
🔗 Read the winner’s paper here!