TIGER: Text-Informed Generalized Enzyme-Reaction Retrieval
Researchers have developed TIGER, a Text-Informed Generalized Enzyme-Reaction Retrieval framework, to address poor generalization in enzyme-reaction retrieval tasks. The framework uses protein-to-text generation models to extract textual semantic knowledge from enzyme sequences, creating a generalized representation that links enzymes and biochemical reactions. A Dynamic Gating Network adaptively fuses this text-derived knowledge to improve retrieval reliability. The work is published on arXiv (2605.24489).
Key facts
- TIGER stands for Text-Informed Generalized Enzyme-Reaction Retrieval.
- It addresses enzyme-reaction retrieval, a bidirectional task mapping enzymes to reactions and vice versa.
- Existing approaches suffer from poor generalization and asymmetry between retrieval directions.
- TIGER uses protein-to-text generation models to distill textual semantics from enzyme sequences.
- A Dynamic Gating Network adaptively fuses text-derived knowledge.
- The paper is available on arXiv with ID 2605.24489.
- The framework aims to improve performance across tasks and distributions.
- It supports enzyme characterization, reaction mechanism elucidation, and metabolic pathway design.
Entities
Institutions
- arXiv