Research map / Reinforcement learning / RL from feedback
FIDIA: Function-Informed Sequence Design via Inference-Aligned Policy Optimization
Minghan Li, Fengji Li, Yilin Tao, Yue Deng
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 expectation objective with Best-of-N rank-based policy gradient (FIDIA) on ProteinMPNN | ProteinMPNN · Motif scaffolding FrameF | Success rate (%): 26.67 -> 37.50 | Improved |
| replace expectation objective with Best-of-N rank-based policy gradient (FIDIA) on ProteinMPNN | ProteinMPNN · Motif scaffolding RFD | Success rate (%): 55.00 -> 65.83 | Improved |
| use FIDIA instead of GRPO with standard expected-return objective | ProteinMPNN · Motif scaffolding FrameF | Success rate (%): 33.33 -> 37.50 | Improved |
| BoN baseline trained on top-k candidates instead of top-1 | ProteinMPNN · Motif scaffolding FrameF | Success rate (%): 34.17 -> 28.33 | Hurt |
| apply FIDIA instead of MPNN: diversity of sampled sequences | ProteinMPNN · Motif scaffolding RFD | Diversity: 0.31 -> 0.43 | Improved |
| cross-validate with AlphaFold2 instead of ESMFold oracle | FIDIA · Motif scaffolding FrameF | Success rate (%): 37.50 -> 36.67 | No clear effect |
| apply FIDIA to epitope scaffolding vs MPNN | ProteinMPNN · Epitope scaffolding (vaccine) | Success rate (%): 25.45 -> 27.09 | Improved |
| fine-tune LigandMPNN with FIDIA using Vina reward | LigandMPNN · Affinity-enhancing enzyme design | Vina score (kcal/mol): -6.53 -> -7.21 | Improved |
| fine-tune LigandMPNN with FIDIA using Vina reward | LigandMPNN · Affinity-enhancing enzyme design | Success rate (%): 90.74 -> 83.33 | Hurt |
| increase structure sampling budget M to 50 (ablation) | FIDIA · Motif scaffolding (pipeline) | Solved motifs: 21 targets at M=50 | Improved |
Closest papers by meaning
- ProteinOPD: Towards Effective and Efficient Preference Alignment for Protein Design (NeurIPS 2026 · similarity 0.86)
- Sidechain conditioning and modeling for full-atom protein sequence design with FAMPNN (ICML 2025 · similarity 0.86)
- Improved Off-policy Reinforcement Learning in Biological Sequence Design (ICML 2025 · similarity 0.86)
- ProSpero: Active Learning for Robust Protein Design Beyond Wild-Type Neighborhoods (NeurIPS 2025 · similarity 0.86)
- Protein Design with Dynamic Protein Vocabulary (NeurIPS 2025 · similarity 0.86)
- Steering Generative Models with Experimental Data for Protein Fitness Optimization (NeurIPS 2025 · similarity 0.86)
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.
