Reaction center prediction by analyzing attention of a chemical language model
Abstract
This study proposes PABA and SPABA to predict reaction centers by interpreting attention matrices of CLMs. The 7_7 attention head is optimal; SPABA_7_7 reaches an MCC of 0.73, outperforming existing supervised methods with strong generalizability.
Fields of Study & Subjects
Machine Learning in Materials Science
Computational Drug Discovery Methods
Protein Structure and Dynamics
Publication details
- Journal / source
- Digital Discovery
- Publisher
- Royal Society of Chemistry (RSC)
- Volume
- 5
- Issue
- 6
- Pages
- 2458–2468
- Publication date
- 1 January 2026
- Language
- EN