Property Optimization on MOSES ~1.5M molecules (test)
85C.V.U.N. ScoreGEM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GEMInitialization=Noise Initialization, Condition=logP <= 1.5, Inference steps=1k, Evaluation samples=5k2026.03 | 85 | 85 | |
| GEMInitialization=Noise Initialization, Condition=logS >= -2.25, Inference steps=1k, Evaluation samples=5k2026.03 | 80 | 81 | |
| GEMInitialization=Noise Initialization, Condition=TPSA <= 50, Inference steps=1k, Evaluation samples=5k2026.03 | 80 | 81 | |
| DeFoGInitialization=Noise Initialization, Condition=logP <= 1.5, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 78 | 82 | |
| DeFoGInitialization=Noise Initialization, Condition=TPSA <= 50, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 77 | 82 | |
| DeFoGInitialization=Noise Initialization, Condition=logS >= -2.25, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 74 | 79 | |
| G-VFMInitialization=Noise Initialization, Condition=logS >= -2.25, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 72 | 75 | |
| G-VFMInitialization=Noise Initialization, Condition=TPSA <= 50, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 72 | 80 | |
| G-VFMInitialization=Noise Initialization, Condition=logP <= 1.5, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 70 | 74 | |
| GEMInitialization=Noise Initialization, Condition=QED >= 0.9, Inference steps=1k, Evaluation samples=5k2026.03 | 40 | 40 | |
| G-VFMInitialization=Noise Initialization, Condition=QED >= 0.9, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 26 | 28 | |
| DeFoGInitialization=Noise Initialization, Condition=QED >= 0.9, reproduced=true, Inference steps=1k, Evaluation samples=5k2026.03 | 26 | 31 | |
| Training samples (MOSES)Condition=logS >= -2.25, Inference steps=1k, Evaluation samples=5k2026.03 | 0 | 24 | |
| Training samples (MOSES)Condition=QED >= 0.9, Inference steps=1k, Evaluation samples=5k2026.03 | 0 | 16 | |
| Training samples (MOSES)Condition=TPSA <= 50, Inference steps=1k, Evaluation samples=5k2026.03 | 0 | 21 | |
| Training samples (MOSES)Condition=logP <= 1.5, Inference steps=1k, Evaluation samples=5k2026.03 | 0 | 15 |