The abstract is read at the publisher; the summary is JClub's.
EnzRef predicts enzyme turnover numbers kcat from protein sequences and reaction chemistry, improving mean R² by 19.8% on unseen enzymes versus MCKcat without target-domain adaptation.
- Why it matters: Virtual screening needs reliable ranking across enzyme–reaction pairs, but existing multimodal models blur reactant-versus-product roles and mix enzyme-specific background with chemistry-dependent effects.
- What they did: EnzRef adapts frozen protein language models using reference-centered cross-attention that decomposes chemistry into reactant and product conditions, then recomposes role-specific deviations into an activity representation.
- The result: With 1.72M trainable parameters beyond the frozen encoder, EnzRef reaches state-of-the-art source-domain prediction on CatPred-DB and zero-shot transfer to TurNuP-DB.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.