Research map / Multi-agent systems and games / Cooperative multi-agent RL
CocoRNA: Collective RNA Design with Cooperative Multi-agent Reinforcement Learning
Tianmeng Hu, Biao Luo, Ke 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 |
|---|---|---|---|
| Add multi-agent decomposition (4 agents) instead of a single agent (ablation) | CocoRNA · Rfam RNA inverse design (650 test structures) | success rate (%): 52.43 -> 97.75 | Improved |
| Add search-augmented exploration (SAE) (ablation) | CocoRNA · Rfam RNA inverse design (650 test structures) | success rate (%): 92.25 -> 97.75 | Improved |
| Add SAE under terminal-only reward (ablation) | CocoRNA · Rfam, terminal reward | success rate (%): 87.69 -> 92.92 | Improved |
| Add SAE under delayed reward (ablation) | CocoRNA · Rfam, delayed reward | success rate (%): 95.54 -> 97.54 | Improved |
| Use position-based instead of random decomposition (ablation) | CocoRNA · Rfam, six splits | success rate (%): 91.74 -> 97.44 | Improved |
| Use position-based instead of structure-type decomposition (ablation) | CocoRNA · Rfam, six splits | success rate (%): 96.69 -> 97.44 | No clear effect |
| Increase agents from 2 to 4 (ablation) | CocoRNA · Rfam RNA inverse design (650 test structures) | success rate (%): 92.46 -> 97.85 | Improved |
| Increase agents from 4 to 8 (ablation) | CocoRNA · Rfam RNA inverse design (650 test structures) | success rate (%): 97.85 -> 94.62 | Hurt |
| Share policy parameters across agents (ablation) | CocoRNA-PBD · Rfam RNA inverse design (650 test structures) | success rate (%): 97.85 -> 95.69 | Hurt |
| Use terminal-only reward instead of per-step reward (ablation) | CocoRNA · Rfam RNA inverse design (650 test structures) | success rate (%): 97.75 -> 92.92 | Hurt |
| Use delayed reward every 10 steps instead of every step (ablation) | CocoRNA · Rfam RNA inverse design (650 test structures) | success rate (%): 97.75 -> 97.54 | No clear effect |
| Use CocoRNA instead of single-agent on TF Bind 10 and ChEMBL (ablation) | CocoRNA · DesignBench TF Bind 10 / ChEMBL | score: 0.725 -> 1.000; 0.595 -> 0.811 | Improved |
Closest papers by meaning
- TANGO: RNA Topology and Geometry Co-Design (NeurIPS 2026 · similarity 0.91)
- RNAFlow: RNA Structure & Sequence Design via Inverse Folding-Based Flow Matching (ICML 2024 · similarity 0.87)
- RDesign: Hierarchical Data-efficient Representation Learning for Tertiary Structure-based RNA Design (ICLR 2024 · similarity 0.86)
- gRNAde: Geometric Deep Learning for 3D RNA inverse design (ICLR 2025 · similarity 0.86)
- RiboFlow: Conditional De Novo RNA Co-Design via Synergistic Flow Matching (NeurIPS 2025 · similarity 0.85)
- Searching for High-Value Molecules Using Reinforcement Learning and Transformers (ICLR 2024 · similarity 0.84)
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.
