Public defence in Computer Science, M.Sc. Maryam Astero

Title of the thesis: Deep Learning for Chemical Reactions

Thesis defender: Maryam Astero
Opponent: Professor Giorgio Valentini, University of Milan, Italy
Custos: Professor Juho Rousu, Aalto University School of Science

Chemical reactions are central to molecular design, materials development, and drug discovery. Understanding how atoms rearrange and how bonds change during reactions is essential, yet existing computational models often struggle to generalize beyond simple or well-studied reactions.

This thesis investigates how deep learning can model chemical reactions more accurately and interpretably by representing molecules as graphs, with atoms as nodes and bonds as edges. This representation enables neural networks to learn how molecular structures transform during reactions.

The research introduces graph-based learning models that integrate chemical knowledge with data-driven representation learning. These models identify which regions of a molecule are most reactive and how atomic rearrangements occur during chemical transformations. By learning directly from large datasets of chemical reactions, the models can also organize reactions into broader categories and uncover general patterns governing molecular reactivity.

Overall, the thesis demonstrates how deep learning methods can be used to model molecular transformations in an interpretable and scientifically rigorous way. The results contribute to the growing field of computational chemistry and support future advances in reaction prediction, synthesis design, and data-driven molecular discovery.

Keywords: Graph Neural Networks, Deep Graph Matching, Multitask Learning, Symmetry-Aware Learning, Graph Representation Learning, Atom Mapping, Reaction Center Identification,  Chemical Reaction Prediction.

Thesis available for public display 7 days prior to the defence at Aaltodoc. 

Contact information: maryam.astero@aalto.fi 

Zoom Link