Graph neural networks for molecules and processes

Graph neural networks (GNNs) learn directly from graph representations of molecules and processes. We use them to predict molecular properties, to design molecules by optimization, and to model process plants from their topology.

Challenge

Predicting molecular properties is an important and challenging task for which quantitative structure-property relationship (QSPR) models are frequently used. QSPR models require the selection of molecular descriptors or structural groups by hand. Similarly, process data-driven models usually ignore how the units of a plant are connected.

Our approach

GNNs use a graph representation of molecules, where atoms correspond to nodes and bonds to edges. They learn physicochemical properties as a function of the molecular graph in a supervised, end-to-end manner: graph convolutions learn molecular fingerprints, which are then mapped to properties by deep learning. This eliminates the need to select descriptors by hand.

This line of work originates from Artur M. Schweidtmann's PhD research at RWTH Aachen University, where GNNs were developed for fuel ignition quality and fuel design. At TU Delft, we have extended it in several directions:

  • Physically motivated GNN architectures, such as physical pooling functions for molecular property prediction.
  • Optimization over trained GNNs for computer-aided molecular design, using mixed-integer formulations.
  • Environmental impact prediction of chemicals directly from molecular graphs.
  • Process topologies as graphs: GNN soft sensors that use the flowsheet structure of a plant, for example for multistage spray drying with Danone, and GNNs inside reinforcement learning agents for process design.

The GNN software tool for fuel property prediction is available open source.

Key publications

  • Gao, Q., Schulze Balhorn, L., Laera, A., Meys, R., Goßen, J., Weber, J. M., Wernet, G., & Schweidtmann, A. M. (2026). Environmental impacts prediction using graph neural networks on molecular graphs. Computers & Chemical Engineering, 204, 109362. doi:10.1016/j.compchemeng.2025.109362
  • Theisen, M. F., Meesters, G. M. H., & Schweidtmann, A. M. (2026). Graph neural networks for soft sensors: Learning from process topology and operational data. Computers & Chemical Engineering, 206, 109532. doi:10.1016/j.compchemeng.2025.109532
  • McDonald, T., Tsay, C., Schweidtmann, A. M., & Yorke-Smith, N. (2024). Mixed-integer optimisation of graph neural networks for computer-aided molecular design. Computers & Chemical Engineering, 185, 108660. doi:10.1016/j.compchemeng.2024.108660 · Details
  • Rittig, J. G., Gao, Q., Dahmen, M., Mitsos, A., & Schweidtmann, A. M. (2023). Graph neural networks for the prediction of molecular structure–property relationships. In Machine Learning and Hybrid Modelling for Reaction Engineering, Royal Society of Chemistry, 159–181. doi:10.1039/BK9781837670178-00159 · Details
  • Schweidtmann, A. M., Rittig, J. G., Weber, J. M., Grohe, M., Dahmen, M., Leonhard, K., & Mitsos, A. (2023). Physical pooling functions in graph neural networks for molecular property prediction. Computers & Chemical Engineering, 172, 108202. doi:10.1016/j.compchemeng.2023.108202 · Details
  • Rittig, J. G., Ritzert, M., Schweidtmann, A. M., Winkler, S., Weber, J. M., Morsch, P., Heufer, K. A., Grohe, M., Mitsos, A., & Dahmen, M. (2023). Graph machine learning for design of high-octane fuels. AIChE Journal, 69(4), e17971. doi:10.1002/aic.17971 · Details
  • Schweidtmann, A. M., Rittig, J. G., König, A., Grohe, M., Mitsos, A., & Dahmen, M. (2020). Graph neural networks for prediction of fuel ignition quality. Energy & Fuels, 34(9), 11395–11407. doi:10.1021/acs.energyfuels.0c01533 · Details
← All projects