Molecule Generation on MOSES (test)
100ValidityJTN-VAE
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| JTN-VAEModel Category=ARs2024.07 | 100 | 100 | 100 | 0.855 | 0.85 | 91.3 | — | — | — | — | — | — | |
| ChemBFNModel Category=BFNs, Sampling Steps=10, Representation=SELFIES2024.07 | 100 | 100 | 100 | 0.86 | 0.855 | 99.1 | — | — | — | — | — | — | |
| ChemBFNModel Category=BFNs, Sampling Steps=100, Representation=SELFIES2024.07 | 100 | 100 | 100 | 0.848 | 0.842 | 94.7 | — | — | — | — | — | — | |
| ChemBFNModel Category=BFNs, Sampling Steps=1000, Representation=SELFIES2024.07 | 100 | 100 | 100 | 0.847 | 0.841 | 94 | — | — | — | — | — | — | |
| CombinatorialEvaluation Protocol=Unconditional2025.04 | 100 | — | — | 0.873 | 0.867 | 98.8 | 99.1 | — | — | — | — | — | |
| JT-VAEEvaluation Protocol=Unconditional2025.04 | 100 | — | — | 0.855 | 0.849 | 91.4 | 99.9 | — | — | — | — | — | |
| JT-VAEMethod architecture=Graph-Based, Number of samples=10,0002026.06 | 100 | — | — | 0.855 | 0.849 | 91.4 | 99.9 | — | — | — | — | — | |
| BioMatrix-4BMethod architecture=Sequence-Based, Molecular representation=SELFIES, Number of samples=10,0002026.06 | 99.8 | — | — | 0.868 | 0.861 | 92.5 | 100 | — | — | — | — | — | |
| HOG-Diff2025.02 | 99.7 | — | — | — | — | 89.1 | 100 | 96.7 | 0.94 | 0.52 | 8.6 | — | |
| HyformerEvaluation Protocol=Unconditional, Sampling Temperature (τ)=0.92025.04 | 99.6 | — | — | 0.851 | 0.845 | 70.1 | 100 | — | — | — | — | — | |
| GraphGPT-1WEvaluation Protocol=Few-Shot, Scaling Factor (s)=0.252025.04 | 99.5 | — | — | 0.854 | 0.85 | 25.5 | 99.5 | — | — | — | — | — | |
| BioMatrix-1.7BMethod architecture=Sequence-Based, Molecular representation=SELFIES, Number of samples=10,0002026.06 | 99.5 | — | — | 0.865 | 0.858 | 89.1 | 100 | — | — | — | — | — | |
| MolGPTModel Category=ARs2024.07 | 99.4 | 100 | 100 | 0.857 | 0.851 | 79.7 | — | — | — | — | — | — | |
| MolGPTEvaluation Protocol=Unconditional2025.04 | 99.4 | — | — | 0.857 | 0.851 | 79.7 | 100 | — | — | — | — | — | |
| GraphGPT-1WEvaluation Protocol=Few-Shot, Scaling Factor (s)=0.52025.04 | 99.3 | — | — | 0.856 | 0.848 | 33.4 | 99.6 | — | — | — | — | — | |
| HyformerEvaluation Protocol=Unconditional, Sampling Temperature (τ)=1.02025.04 | 99.1 | — | — | 0.856 | 0.85 | 74.9 | 100 | — | — | — | — | — | |
| OMG-GPTMethod architecture=Sequence-Based, Number of samples=10,0002026.06 | 99 | — | — | 0.855 | 0.849 | 90.3 | 100 | — | — | — | — | — | |
| MolGPTMethod architecture=Sequence-Based, Number of samples=10,0002026.06 | 98.8 | — | — | 0.855 | 0.849 | 82.1 | 100 | — | — | — | — | — | |
| HyformerEvaluation Protocol=Unconditional, Sampling Temperature (τ)=1.12025.04 | 98.6 | — | — | 0.861 | 0.855 | 79.1 | 100 | — | — | — | — | — | |
| BioMatrix-4BMethod architecture=Sequence-Based, Molecular representation=SMILES, Number of samples=10,0002026.06 | 98.3 | — | — | 0.867 | 0.86 | 95.4 | 100 | — | — | — | — | — | |
| GraphGPT-1WEvaluation Protocol=Few-Shot, Scaling Factor (s)=1.02025.04 | 97.8 | — | — | 0.86 | 0.857 | 87.1 | 99.7 | — | — | — | — | — | |
| VAEEvaluation Protocol=Unconditional2025.04 | 97.7 | — | — | 0.856 | 0.85 | 69.5 | 99.8 | — | — | — | — | — | |
| VAEMethod architecture=Sequence-Based, Number of samples=10,0002026.06 | 97.7 | — | — | 0.856 | 0.85 | 70 | 99.8 | — | — | — | — | — | |
| CharRNNEvaluation Protocol=Unconditional2025.04 | 97.5 | — | — | 0.856 | 0.85 | 84.2 | 99.9 | — | — | — | — | — | |
| CharRNNMethod architecture=Sequence-Based, Number of samples=10,0002026.06 | 97.5 | — | — | 0.856 | 0.85 | 84.2 | 99.9 | — | — | — | — | — | |
| GraphGPT-1WEvaluation Protocol=Few-Shot, Scaling Factor (s)=2.02025.04 | 97.2 | — | — | 0.85 | 0.847 | 100 | 100 | — | — | — | — | — | |
| GraphINVENTModel Category=ARs2024.07 | 96.4 | 100 | 99.8 | 0.857 | 0.851 | — | — | — | — | — | — | — | |
| GraphINVENT2025.02 | 96.4 | — | — | — | — | — | 99.8 | 95 | 1.22 | 0.54 | 12.7 | — | |
| GraphINVENTMethod architecture=Graph-Based, Number of samples=10,0002026.06 | 96.4 | — | — | 0.857 | 0.851 | — | 99.8 | — | — | — | — | — | |
| BioMatrix-1.7BMethod architecture=Sequence-Based, Molecular representation=SMILES, Number of samples=10,0002026.06 | 95.1 | — | — | 0.873 | 0.866 | 94.1 | 100 | — | — | — | — | — | |
| AEEEvaluation Protocol=Unconditional2025.04 | 93.7 | — | — | 0.856 | 0.85 | 79.3 | 99.7 | — | — | — | — | — | |
| AAEMethod architecture=Sequence-Based, Number of samples=10,0002026.06 | 93.6 | — | — | 0.855 | 0.85 | 79.3 | 100 | — | — | — | — | — | |
| DeFoG2025.02 | 92.8 | — | — | — | — | 92.1 | 99.9 | 98.9 | 1.95 | 0.55 | 14.4 | — | |
| ChemBFNModel Category=BFNs, Sampling Steps=1000, Representation=SMILES2024.07 | 91.6 | 100 | 99.8 | 0.836 | 0.83 | 88 | — | — | — | — | — | — | |
| ChemBFNModel Category=BFNs, Sampling Steps=100, Representation=SMILES2024.07 | 91.1 | 100 | 99.8 | 0.837 | 0.831 | 88.4 | — | — | — | — | — | — | |
| Cometh2025.02 | 90.5 | — | — | — | — | 92.6 | 99.9 | 99.1 | 1.27 | 0.54 | 16 | — | |
| LatentGANModel Category=ARs2024.07 | 89.7 | 100 | 99.7 | 0.857 | 0.851 | 95 | — | — | — | — | — | — | |
| LatentGANEvaluation Protocol=Unconditional2025.04 | 89.7 | — | — | 0.857 | 0.85 | 94.9 | 99.7 | — | — | — | — | — | |
| LatentGANMethod architecture=Sequence-Based, Number of samples=10,0002026.06 | 89.7 | — | — | 0.857 | 0.85 | 94.9 | 99.7 | — | — | — | — | — | |
| DisCo2025.02 | 88.3 | — | — | — | — | 97.7 | 100 | 95.6 | 1.44 | 0.5 | 15.1 | — | |
| DiGressModel Category=DMs2024.07 | 85.7 | — | 100 | — | — | 95 | — | — | — | — | — | — | |
| DiGress2025.02 | 85.7 | — | — | — | — | 95 | 100 | 97.1 | 1.19 | 0.52 | 14.8 | — | |
| DigressMethod architecture=Graph-Based, Number of samples=10,0002026.06 | 85.7 | — | — | 0.855 | 0.849 | 93.6 | 100 | — | — | — | — | — | |
| ChemBFNModel Category=BFNs, Sampling Steps=10, Representation=SMILES2024.07 | 83.5 | 100 | 99.9 | 0.851 | 0.844 | 92.1 | — | — | — | — | — | — | |
| Sc2MolMethod architecture=Sequence-Based, Number of samples=10,0002026.06 | 63.1 | — | — | 0.866 | 0.872 | 98.6 | 99 | — | — | — | — | — | |
| NGramEvaluation Protocol=Unconditional2025.04 | 23.8 | — | — | 0.874 | 0.864 | 96.9 | 92.2 | — | — | — | — | — | |
| HMMEvaluation Protocol=Unconditional2025.04 | 7.6 | — | — | 0.847 | 0.81 | 99.9 | 56.7 | — | — | — | — | — | |
| DeFoG*Initialization=Noise, Generations=25k, Inference steps=10002026.03 | — | — | — | — | — | — | — | — | 1.95 | — | — | 82.2 | |
| GEMInitialization=Noise, Generations=25k, Inference steps=10002026.03 | — | — | — | — | — | — | — | — | 1.51 | — | — | 85.6 | |
| GEMInitialization=Data, Generations=25k, Inference steps=10002026.03 | — | — | — | — | — | — | — | — | 0.76 | — | — | 89.8 | |
| GraphEBM*Initialization=Noise, Generations=25k, Inference steps=10002026.03 | — | — | — | — | — | — | — | — | 9.83 | — | — | 8.1 | |
| GraphEBM*Initialization=Data, Generations=25k, Inference steps=10002026.03 | — | — | — | — | — | — | — | — | 7.31 | — | — | 32.2 | |
| Training samples (MOSES)Generations=25k, Inference steps=10002026.03 | — | — | — | — | — | — | — | — | 0.25 | — | — | 0 | |
| VFM*Initialization=Noise, Generations=25k, Inference steps=10002026.03 | — | — | — | — | — | — | — | — | 2.71 | — | — | 81.4 |