Machine learning in chemical reaction space
Problem
Chemical reaction space is vastly larger than compound space, and existing machine-learning concepts could not directly handle its hub-based topology.
Why it matters
Predicting reaction energies could accelerate the study of chemical transformations such as combustion.
What the AI did
Machine learning was adapted to reaction space using a new first-principles database (Rad-6 and Rad-6-RE) and modified compound-space ML concepts, demonstrated on methane combustion.
Results
Established Rad-6 and Rad-6-RE databases; demonstrated learned reaction energies for methane combustion (quantitative figures truncated).
Impact
Shows ML can offer a non-empirical route to reaction energies, opening reactive ML research.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Sina Stocker, Gábor Cśanyi, Karsten Reuter, Johannes T. Margraf |
| Citations | 184 (OpenAlex, at import time) |
| References | 78 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-020-19267-x |
| Open access | gold |
| Keywords | machine learning, reaction space, reaction energies, methane combustion, chemical databases |
Importance rating
significant — thousands to millions affected
Opens ML study of reaction space, foundational to chemistry as a science of transformations; abstract quantifies compound space at 10^60 molecules and positions reaction space as even larger.