Reaction informatics and graph-based molecular transformations

This direction develops graph-based and mathematically structured representations for chemical transformations, including reaction curation, rebalancing, graph-based rule extraction, partial ITS formulations, and benchmark-oriented datasets. The main scientific aim is to preserve chemical meaning while enabling scalable analysis and machine learning.

Key themes

• reaction curation and rebalancing

• graph-based reaction rules and template extraction

• chemically faithful reaction representations

• benchmark construction and dataset quality

• emerging work toward mechanistic transition graphs and electron-flow reasoning

Representative outputs include work on reaction rebalancing, SynTemp, SynKit, formal studies of partial ITS graphs, and the SynRXN benchmark. Together, these contributions aim to make reaction data more reliable, more interpretable, and more reusable for downstream modeling.

Why it matters

Reaction data are only as useful as their representations and curation quality. Structured reaction modeling makes it easier to integrate heterogeneous data, train interpretable models, and create reusable knowledge bases for synthesis planning, mechanistic reasoning, and molecular design.