Predicting polymerization reactions via transfer learning using chemical language models
Creators
- 1. IBM Research Brazil - Avenida República do Chile, 330 - 11o. e 12. andares Rio De Janeiro, RJ 20031-170, Brazil
- 2. IBM Research Europe - Säumerstrasse 4, 8803 Rüschlikon, Switzerland
- 3. National Center for Competence in Research-Catalysis (NCCR-Catalysis), Switzerland
* Contact person
Description
Polymers are candidate materials for a wide range of sustainability applications such as carbon capture and energy storage. However, computational polymer discovery lacks automated analysis of reaction pathways and stability assessment through retro-synthesis. Here, we report the first extension of transformer-based language models to polymerization reactions for both forward and retrosynthesis tasks. We curated a polymerization dataset for vinyl polymers covering reactions and retrosynthesis for representative homo-polymers and co-polymers. Overall, we report a forward model accuracy of 80% and a backward model accuracy of 60%. We further analyse the model performance on a set of case studies by providing polymerization and retro-synthesis examples and evaluating the model's predictions quality from a materials science perspective.