Research map / Probabilistic and scientific modeling / Molecules and proteins
ProteinInvBench: Benchmarking Protein Inverse Folding on Diverse Tasks, Models, and Metrics
Zhangyang Gao, Cheng Tan, Yijie Zhang, Xingran Chen, Lirong Wu, Stan Z. Li
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 |
|---|---|---|---|
| use CATH4.3 instead of CATH4.2 for evaluation | single-chain models · CATH | recovery | No clear effect |
| increase protein length | graph-based inverse folding models · CATH4.2/4.3 | recovery | Improved |
| use KWDesign over PiFold (retrieval/refinement with pretrained models) | KWDesign · CATH4.2 | recovery: 0.52 -> 0.61 | Improved |
| evaluate on multi-chain PDB instead of CATH | inverse folding models · PDB | recovery: 0.58 -> 0.66 | Improved |
| train on PDB instead of CATH for de novo evaluation | ProteinMPNN · TS45 | recovery: 0.44 -> 0.52 | Improved |
| train on PDB instead of CATH for de novo evaluation | PiFold · TS45 | recovery: 0.47 -> 0.53 | Improved |
| evaluate on TBM-easy vs TBM-hard subsets | all models · TS45 | recovery | Hurt |
| raise sampling temperature from 0.0 to 0.1 | models pre-trained on CATH4.3 · TS45 | diversity | Improved |
| raise sampling temperature to 0.5 | models pre-trained on CATH4.3 · TS45 | recovery and sc-TM | Hurt |
| use stronger model at same recovery level | inverse folding models · TS45 | diversity | Improved |
| add Gaussian noise epsilon=1 to input structures | KWDesign, PiFold, ProteinMPNN, AlphaDesign · CATH4.3 | recovery: at least 20% decrease | Hurt |
| add Gaussian noise epsilon=1 to input structures | StructGNN and GraphTrans · CATH4.3 | recovery: 14% decrease | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PiFold — PiFold: Toward effective and efficient protein inverse folding (ICLR 2023) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ProteinBench: A Holistic Evaluation of Protein Foundation Models (ICLR 2025 · similarity 0.91)
- ProtInvTree: Deliberate Protein Inverse Folding with Reward-guided Tree Search (NeurIPS 2025 · similarity 0.89)
- Bridge-IF: Learning Inverse Protein Folding with Markov Bridges (NeurIPS 2024 · similarity 0.88)
- Protein Inverse Folding From Structure Feedback (NeurIPS 2025 · similarity 0.88)
- All-atom inverse protein folding through discrete flow matching (ICML 2025 · similarity 0.87)
- Fast Uncovering of Protein Sequence Diversity from Structure (ICLR 2025 · similarity 0.87)
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.
