Peptide SMILES Generation
98.7Validity ScoreSFT-RL
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SFT-RLNumber of samples=502025.08 | 98.7 | 99.8 | 23.8 | 83.3 | 9.84 | 95.6 | |
| SFT-RLNumber of samples=102025.08 | 98 | 99.9 | 51 | 83 | 4.12 | 93.7 | |
| CoT-SFTNumber of samples=502025.08 | 95.6 | 99.9 | 95.4 | 19.6 | 8.82 | 87.9 | |
| CoT-SFTNumber of samples=102025.08 | 95.3 | 99.9 | 98.9 | 19.7 | 1.85 | 67.1 | |
| GPT-4oNumber of shots=1-shot2025.08 | 92.6 | 99.6 | 74.6 | 2.3 | 0.17 | 7 | |
| PepThink-R1training_stage=CoT-SFT-RL, Number of samples=102025.08 | 90 | 100 | 56 | 89 | 4.42 | 96 | |
| PepThink-R1training_stage=CoT-SFT-RL, Number of samples=502025.08 | 90 | 100 | 30 | 89 | 11.81 | 98.4 | |
| Random mutation2025.08 | 87.6 | 99.8 | 99.1 | 0.3 | 0.61 | 20.8 | |
| GPT-4oNumber of shots=0-shot2025.08 | 84 | 98.4 | 93.7 | 4.3 | 0.35 | 19.8 | |
| GPT-5reasoning effort=minimal2025.08 | 82.1 | 99.6 | 95.9 | 10.7 | 0.85 | 38.6 | |
| DeepSeek-R1-Llama-8BNumber of shots=1-shot2025.08 | 67.9 | 87.5 | 42.8 | 4.1 | 0.12 | 7.9 | |
| DeepSeek-R1-Llama-8BNumber of shots=0-shot2025.08 | 55.7 | 83.9 | 55.5 | 3.3 | 0.12 | 7.2 | |
| Qwen3-30B-thinkingNumber of shots=0-shot2025.08 | 37.4 | 95.4 | 36.8 | 5.2 | 0.04 | 3.2 | |
| Qwen3-30B-thinkingNumber of shots=1-shot2025.08 | 34.6 | 98.9 | 39 | 1.5 | 0.02 | 1.8 |