Research map / Probabilistic and scientific modeling / Molecules and proteins
Proximal Exploration for Model-guided Protein Sequence Design
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 |
|---|---|---|---|
| Proximal exploration (PEX) with MuFacNet instead of AdaLead-style CNN model-guided evolution | model-guided protein sequence design · 8 protein tasks (overall normalized) | Overall normalized max fitness: 0.98 vs 0.43 (AdaLead + CNN) | Helped |
| Replace MuFacNet with CNN surrogate in PEX (ablation) | PEX · 8 protein tasks (overall normalized) | Overall normalized max fitness: 0.89 vs 0.98 | Hurt |
| Use MuFacNet surrogate instead of CNN (ablation) | AdaLead · 8 protein tasks (overall normalized) | Overall normalized max fitness: 0.78 vs 0.43 | Helped |
| Use MuFacNet surrogate instead of CNN (ablation) | DbAS and CbAS · 8 protein tasks (overall normalized) | Overall normalized max fitness: DbAS 0.56 vs 0.36; CbAS 0.58 vs 0.38 | Helped |
| Use MuFacNet surrogate instead of CNN in DyNA PPO (ablation) | DyNA PPO · 8 protein tasks (overall normalized) | Overall normalized max fitness: 0.05 vs 0.09 | No clear effect |
| Use context window (radius 10) encoding vs window radius 5 or 15 for mutation sites (ablation) | MuFacNet · avGFP, AAV, TEM, E4B | Max fitness after 10 rounds: avGFP 3.12 (r10) vs 2.96 (r5) vs 2.86 (r15); AAV 4.45 vs 4.68 vs 4.58 | Mixed |
| Replace context window encoding with positional encoding of mutations (ablation) | MuFacNet · avGFP, AAV, TEM, E4B | Max fitness after 10 rounds: avGFP 3.15 vs 3.12; AAV 4.32 vs 4.45 | No clear effect |
| Reject sequences with more than 10 mutations (RD) added to AdaLead (ablation) | AdaLead · AAV | Max fitness after 10 rounds: -0.12 vs 2.52 for PEX | Hurt |
| Add lower confidence bound (LCB) or distance rejection to AdaLead/CbAS (ablation) | AdaLead, CbAS · avGFP, AAV, TEM, E4B | Max fitness after 10 rounds: e.g., AdaLead+LCB avGFP 2.67 vs PEX 2.97; AAV -2.30 vs 2.52 | Mixed |
| PEX proximal exploration with low mutation count vs greedy baselines | PEX · Mutation count of best designed sequence across benchmarks | Mutation count: qualitative: far fewer mutations than baselines | Helped |
Closest papers by meaning
- ProSpero: Active Learning for Robust Protein Design Beyond Wild-Type Neighborhoods (NeurIPS 2025 · similarity 0.90)
- Tree Search-Based Evolutionary Bandits for Protein Sequence Optimization (AAAI 2024 · similarity 0.89)
- A Variational Perspective on Generative Protein Fitness Optimization (ICML 2025 · similarity 0.88)
- Steering Generative Models with Experimental Data for Protein Fitness Optimization (NeurIPS 2025 · similarity 0.88)
- Knowledge-aware Reinforced Language Models for Protein Directed Evolution (ICML 2024 · similarity 0.87)
- DePLM: Denoising Protein Language Models for Property Optimization (NeurIPS 2024 · 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.