Research map / Vision-language and multimodal / Cross-modal learning
PoinnCARE: Hyperbolic Multi-Modal Learning for Enzyme Classification
Kun XIE, Peng Zhou, Xingyi Zhang, Wei Liu, Peilin Zhao, Sibo Wang, Biaobin Jiang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Euclidean MLP with hyperbolic embedding (ablation) | MLP classifier · CARE <30% Identity | level-4 accuracy: 9.3% | Improved |
| add modality alignment fusing active site and structure (ablation) | PoinnCARE with structure only · CARE <30% Identity | level-4 accuracy: 2.6% | Improved |
| add active site graph or structural similarity graph (ablation) | hyperbolic MLP · CARE <30% Identity | accuracy | Improved |
| move ESM2 embeddings to hyperbolic space (ablation) | ESM2 · CARE <30% Identity | level-4 accuracy: 10.6% | Improved |
| move ProtT5 embeddings to hyperbolic space (ablation) | ProtT5 · CARE <30% Identity | level-4 accuracy: 11.8% | Improved |
| move ESMc embeddings to hyperbolic space (ablation) | ESMc · CARE <30% Identity | level-4 accuracy: 19.0% | Improved |
| use full PoinnCARE framework on top of hyperbolic encoder (ablation) | ESM2/ProtT5/ESMc · CARE <30% Identity | level-4 accuracy: up to an additional 8.2% | Improved |
| reduce embedding dimension 512 to 32 (ablation) | CLEAN · CARE <30% Identity | level-4 accuracy: 0.535 -> 0.354 | Hurt |
| reduce embedding dimension to 32 (ablation) | PoinnCARE · CARE <30% Identity | level-4 accuracy: 0.597 | No clear effect |
| PoinnCARE vs CLEAN | CLEAN · CARE <30% Identity | level-4 accuracy: 0.535 -> 0.648 | Improved |
| PoinnCARE vs CLEAN | CLEAN · CARE Promiscuous | level-4 accuracy: 9.4% | Improved |
| PoinnCARE vs ESM-2 | ESM-2 · CARE Price | level-4 accuracy: 0.362 -> 0.349 | Hurt |
Closest papers by meaning
- CARE: a Benchmark Suite for the Classification and Retrieval of Enzymes (NeurIPS 2024 · similarity 0.86)
- Evaluating Representation Learning on the Protein Structure Universe (ICLR 2024 · similarity 0.84)
- MMSite: A Multi-modal Framework for the Identification of Active Sites in Proteins (NeurIPS 2024 · similarity 0.84)
- TRIDENT: Tri-Modal Molecular Representation Learning with Taxonomic Annotations and Local Correspondence (NeurIPS 2025 · similarity 0.84)
- Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism (NeurIPS 2024 · similarity 0.83)
- Retrieval Augmented Zero-Shot Enzyme Generation for Specified Substrate (ICML 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
