Esther Heid
co-PI in P09
Institute of Materials Chemistry
TU Wien
Getreidemarkt 9
1060 Vienna, Austria
Research interests
My research is focused on machine learning and deep learning for molecules, biomolecules, materials, and chemical reactions. I am especially interested in uncertainty quantification, i.e. not only increasing the accuracy of a model prediction, but also quantifying how certain or uncertain a specific prediction is. This uncertainty can then in turn be used to train better models using active learning. I am also interested in writing and maintaining software.
Most relevant scientific results
- Published several open-source softwares and databases (Chemprop, EnzymeMap, ESPsim, TemplateCorr, EHreact) including the popular graph-based machine learning software Chemprop for molecular and reaction properties, as well as the current state-of-the-art enzymatic reaction database EnzymeMap.
- Pioneered the development of a spatially-resolved uncertainty approach for machine learning potentials.
- Pioneered the prediction of reaction properties using the condensed graph of reaction (overlay of the reactant and product graphs) as representation.
- Developed a method to determine atomic polarizabilities (needed as input parameters for polarizable molecular dynamics simulations) from quantum mechanical calculations.
Career
- 2013–present: University assistant, Institute for Materials Chemistry, TU Wien, Austria
- 2022–2023: Project assistant, Institute for Materials Chemistry, TU Wien, Austria
- 2020–2022: Postdoctoral fellow, Department of Chemical Engineering, MIT, US
Education
- 2019 PhD in Theoretical Chemistry (Dr. rer. nat.), University of Vienna, Austria
- 2016 MSc in Chemistry, University of Vienna, Austria
- 2014 BSc in Chemistry, University of Vienna, Austria /li>