Research map / Large language models / Language, safety and interpretability
Structure-informed Language Models Are Protein Designers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add LM-Design to the PiFold structure encoder | PiFold · CATH 4.2 all | median recovery (%): 50.22 -> 55.65 | Helped |
| Add LM-Design to the PiFold structure encoder | PiFold · CATH 4.3 all | median recovery (%): 50.68 -> 56.63 | Helped |
| Add the ESM-1b-based LM-Design decoder | ProteinMPNN + CMLM · CATH 4.2 all | median recovery (%): 48.62%→52.99% | Helped |
| Add the ESM-1b-based LM-Design decoder | ProteinMPNN + CMLM · CATH 4.3 all | median recovery (%): 48.25%→54.05% | Helped |
| Replace ESM-1b with ESM-2 3B (ablation) | LM-Design for protein complexes · Multi-chain protein complexes | median recovery (%): 59.10 -> 61.49 | Helped |
| Use a pretrained GVP-TransformerEncoder (ablation) | LM-Design for protein complexes · Multi-chain protein complexes | median recovery (%): 59.10 -> 62.16 | Helped |
| Freeze rather than fine-tune the pretrained ProteinMPNN-CMLM encoder (ablation) | LM-Design · CATH 4.2 all | median recovery (%): 54.62 -> 54.41 | No clear effect |
| Freeze rather than fine-tune the pretrained ProteinMPNN-CMLM encoder (ablation) | LM-Design · CATH 4.3 all | median recovery (%): 55.92 -> 56.38 | Helped |
| Iteratively refine the predicted sequence (ablation) | LM-Design · CATH test evaluation | sequence recovery: 1~2% | Helped |
| Increase AlphaFold 2 predicted-structure augmentation to 20k examples (ablation) | LM-Design and ProteinMPNN+CMLM · CATH evaluation | sequence recovery | Helped |
| Increase predicted-structure augmentation to 50k examples (ablation) | LM-Design · CATH evaluation | sequence recovery | Helped |
| Increase predicted-structure augmentation to 50k examples (ablation) | ProteinMPNN+CMLM · CATH evaluation | sequence recovery | Hurt |
Closest papers by meaning
- Steering Protein Language Models (ICML 2025 · similarity 0.92)
- Elucidating the Design Space of Multimodal Protein Language Models (ICML 2025 · similarity 0.91)
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.91)
- Protein Design with Dynamic Protein Vocabulary (NeurIPS 2025 · similarity 0.90)
- SaProt: Protein Language Modeling with Structure-aware Vocabulary (ICLR 2024 · similarity 0.90)
- From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models (ICML 2025 · similarity 0.90)
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.