"Machine learning in chemistry and beyond" (ChE-651) seminar by Esther Heid "Graph-convolutional neural network: From molecules to chemical reactions"

Graph-convolutional neural networks (GCNNs) are very successful in predicting various properties of molecules, outperforming descriptor or fingerprint based neural networks if plenty of data is available. GCNNs allow for a learned extraction of important characteristics of a molecule and enable end-to-end learning, instead of relying on expert, system-dependent knowledge. However, the properties of chemical reactions, i.e. the combination of reactant and product molecules, are not readily accessible with current GCNNs which are designed to take molecular graphs as input. We therefore developed GCNNs based on the condensed graphs of reaction (CGR), where reactants and products are merged into a single pseudo-molecular graph, i.e. an artificial graph transition state. In this talk, the anatomy of molecular GCNNs will be discussed in detail, as well as the changes necessary to encode reactions instead of molecules. Compared to previous approaches such as neural networks on reaction fingerprints, transformer models on reaction SMILES strings or separate GCNNs for reactant and product encodings, GCNNs on CGRs offer a comparable or better performance with a lower number of parameters. We showcase the performance on different regression and classification tasks, such as the prediction of activation energies, rate constants or the classification into name reactions. Furthermore, possible pitfalls for molecule and reaction GCNNs will be discussed.

Esther Heid received her Bachelor’s (2014) and Master’s degree (2016) in Chemistry from the University of Vienna, Austria, as well as a doctoral degree in Chemistry, with focus on Theoretical Chemistry, in 2019. Within her PhD program she visited the London Imperial College, as well as the University of Maryland Baltimore for short term research trips.
In her thesis, she investigated how computer simulation can be utilized to study solvation dynamics and other solvent effects in bulk phase and close to biomolecules such as saccharides or proteins. Her expertise comprises molecular dynamics simulations, quantum mechanical calculations, machine learning, method development and the automation of work flows.
In 2020 she joined the Green Research group, holding an Erwin-Schroedinger Postdoctoral Fellowship from the Austrian Science Fund, which enables her to conduct research on the development of a computer-aided tool for finding novel multi-enzyme networks which yield a specified target molecule. The project utilizes recent developments in biocatalysis, retrosynthesis and cheminformatics, and aims toward a more efficient, selective and environmentally favorable synthesis of compounds through the inclusion of biocatalytic transformations. She furthermore works on developing new machine learning methods for molecular and reaction property predictions.
Visit LinkedIn, Google Scholar or Github to connect or learn more.