Machine learning has become a powerful tool for predicting and even designing molecular properties, materials, and chemical reactions. By nature, machine learning is data-driven; it aims to uncover complex patterns, such as the solubility or antibiotic activity of a compound, directly from its structure. However, generating large, high-quality chemical datasets is often expensive and time-consuming, making the development of data-efficient models a key focus in the field. Successful predictions typically rely on machine learning architectures that incorporate domain knowledge, such as the chemical and physical symmetries of a system. But how much can such models actually support the day-to-day work of a chemist?
In this talk, I will provide an overview of the field and present my work on modeling molecular properties,[1] reaction properties,[2] retrosynthesis,[3] and enzymatic reactions,[4,5] together with an evaluation of their practical impact and usability. I will also discuss the challenge of quantifying uncertainty in chemical machine learning,[6,7] that is, evaluating the reliability of predictions of a model. Finally, I will discuss my current research efforts aimed at advancing more sustainable chemistry through machine learning, with a focus on predicting chemical reactions with and without catalysts.
[1] E. Heid, K. P. Greenman, Y. Chung, S.-C. Li, D. E. Graff, F. H. Vermeire, H. Wu, W. H. Green, and C. J. McGill. J. Chem. Inf. Model., 2023, 64, 9–17.
[2] E. Heid and W. H. Green. J. Chem. Inf. Model., 2021, 62, 2101–2110.
[3] E. Heid, J. Liu, A. Aude and W. H. Green. J. Chem. Inf. Model., 2021, 62, 16–26.
[4] E. Heid, D. Probst, W. H. Green and G. K. H. Madsen. Chem. Sci., 2023, 14, 14229– 14242.
[5] E. Heid, S. Goldman, K. Sankaranarayanan, C. W. Coley, C. Flamm and W. H. Green. J. Chem. Inf. Model., 2021, 61, 4949–4961.
[6] E. Heid, J. Schorghuber, R. Wanzenbock and G. K. H. Madsen. J. Chem. Inf. Model., 2024, 64, 6377–6387.
[7] E. Heid, C. J. McGill, F. H. Vermeire and W. H. Green. J. Chem. Inf. Model., 2023, 63, 4012–4029.

Einladung, Details und Abstract