Molecular AI and chemistry-aware learning

A second major direction of my research is the development of machine-learning methods for molecular systems, with emphasis on chemically meaningful representations, rigorous evaluation, and the integration of domain knowledge.

Key themes

• graph neural networks for reaction classification

• knowledge-guided learning for molecular property prediction

• dataset design and benchmark-aware evaluation

• interpretable and transferable molecular representations

Representative outputs include SynCat for reaction classification and KGG for knowledge-guided graph learning in molecular property prediction. In both settings, a recurring priority is to combine predictive performance with problem-specific chemical structure and clearer scientific interpretation.

Scientific perspective

I am especially interested in models that do not treat molecules as generic graphs alone, but instead preserve chemically relevant constraints, role information, and transformation context. That perspective helps align molecular AI more closely with scientific use, especially when models are expected to support interpretation, benchmarking, and downstream discovery.