Unconditional Molecule Generation on GuacaMol
100ValidityMOLER
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MOLERYear=20222026.02 | 100 | 100 | 62.5 | 96.4 | 99.1 | — | — | |
| MICAMYear=20232026.02 | 100 | 99.4 | 73.1 | 98.9 | 98.6 | — | — | |
| DISCO-GTYear=20242026.02 | 100 | 86.6 | 59.7 | 92.6 | 99.9 | — | — | |
| Training setNumber of runs=3, Filtering=unfiltered2026.02 | 100 | 100 | 92.8 | 99.9 | — | — | — | |
| Training set2025.06 | 100 | — | 92.8 | — | — | 100 | 0 | |
| JT-VAEInput representation=Graph-based2025.04 | 100 | — | 0.75 | 0.94 | — | — | — | |
| MoLeRInput representation=Graph-based2025.04 | 100 | 100 | 0.625 | 0.964 | 99.1 | — | — | |
| MAGNetInput representation=Graph-based2025.04 | 100 | — | 0.76 | 0.95 | — | — | — | |
| MiCaMInput representation=Graph-based2025.04 | 100 | 99.4 | 0.731 | 0.989 | 98.6 | — | — | |
| BioMatrix-1.7BModel category=Sequence-Based, Molecular representation=SELFIES, Model scale=1.7B2026.06 | 99.9 | 100 | — | — | 97.7 | — | — | |
| BioMatrix-4BModel category=Sequence-Based, Molecular representation=SELFIES, Model scale=4B2026.06 | 99.8 | 100 | — | — | 98.6 | — | — | |
| BioMatrix-1.7BModel category=Sequence-Based, Molecular representation=SMILES, Model scale=1.7B2026.06 | 99.4 | 99.9 | — | — | 95.7 | — | — | |
| BioMatrix-4BModel category=Sequence-Based, Molecular representation=SMILES, Model scale=4B2026.06 | 99.1 | 100 | — | — | 97.8 | — | — | |
| DeFoG2025.06 | 99 | — | 73.8 | — | — | 99 | 97.9 | |
| Cometh2025.06 | 98.9 | — | 72.7 | — | — | 98.9 | 97.6 | |
| BWFlow2025.06 | 98.8 | — | 69.2 | — | — | 98.9 | 97.4 | |
| HyformerSampling temperature τ=0.9, Input representation=SMILES-based, Parameters=8.5M2025.04 | 98.6 | 99.9 | 0.897 | 0.995 | 87.9 | — | — | |
| OMG-GPTModel category=Sequence-Based2026.06 | 98.4 | 100 | — | — | 97.5 | — | — | |
| MolGram (GPT)Params=87.5M2026.06 | 98.29 | 99.46 | 0.9112 | 0.9935 | 98.94 | — | — | |
| MolGram (GPT)Params=27.2M2026.06 | 98.21 | 99.42 | 0.9101 | 0.9931 | 98.92 | — | — | |
| MolGPTParams=85.4M2026.06 | 98.16 | 99.22 | 0.9096 | 0.9929 | 98.96 | — | — | |
| MolGPTInput representation=SMILES-based2025.04 | 98.1 | 99.8 | 0.907 | 0.992 | 100 | — | — | |
| MolGPTParams=25.5M2026.06 | 98.07 | 99.14 | 0.9085 | 0.9924 | 99 | — | — | |
| MolGPTModel category=Sequence-Based2026.06 | 97.9 | 99.9 | — | — | 100 | — | — | |
| HyformerSampling temperature τ=1.0, Input representation=SMILES-based, Parameters=8.5M2025.04 | 97.8 | 99.9 | 0.918 | 0.989 | 90.8 | — | — | |
| HyformerSampling temperature τ=1.1, Input representation=SMILES-based, Parameters=8.5M2025.04 | 96.5 | 100 | 0.894 | 0.977 | 93.1 | — | — | |
| LSTM2026.02 | 95.9 | 100 | 91.3 | 99.1 | 91.2 | — | — | |
| LSTMInput representation=SMILES-based2025.04 | 95.9 | 100 | 0.913 | 0.991 | 91.2 | — | — | |
| SMILES LSTMModel category=Sequence-Based2026.06 | 95.9 | 100 | — | — | 91.2 | — | — | |
| NAGVAEModel category=Graph-Based2026.06 | 92.9 | 95.5 | — | — | 100 | — | — | |
| MolHITmodel=Ours2026.02 | 87.1 | 87.1 | 54.9 | 96.7 | 86 | — | — | |
| MolHITNumber of runs=3, Filtering=unfiltered2026.02 | 87.1 | 87.1 | 54.9 | 96.7 | 86 | — | — | |
| VAEInput representation=SMILES-based2025.04 | 87 | 99.9 | 0.863 | 0.982 | 97.4 | — | — | |
| VAEModel category=Sequence-Based2026.06 | 87 | 99.9 | — | — | 97.4 | — | — | |
| DisCo2025.06 | 86.6 | — | 59.7 | — | — | 86.6 | 86.5 | |
| DiGReSSYear=20232026.02 | 85.2 | 100 | 68 | 92.9 | 99.9 | — | — | |
| DiGressversion=original (trained on filtered dataset)2026.02 | 85.2 | 85.2 | 68 | 92.9 | 85.1 | — | — | |
| DiGress (org.)Number of runs=3, Filtering=unfiltered2026.02 | 85.2 | 85.2 | 68 | 92.9 | 85.1 | — | — | |
| DiGress2025.06 | 85.2 | — | 68 | — | — | 85.2 | 85.1 | |
| DigressModel category=Graph-Based2026.06 | 85.2 | 100 | — | — | 99.6 | — | — | |
| AAEModel category=Sequence-Based2026.06 | 82.2 | 100 | — | — | 99.8 | — | — | |
| DiGressversion=full (trained on unfiltered dataset)2026.02 | 74.7 | 74.6 | 61.1 | 92.4 | 74 | — | — | |
| DiGress (full)Number of runs=3, Filtering=unfiltered2026.02 | 74.7 | 74.6 | 61.1 | 92.4 | 74 | — | — | |
| DiGress + DAEarchitecture=original graph transformer with DAE2026.02 | 65.2 | 65.2 | 49.2 | 87 | 64.9 | — | — | |
| DiGress+DAENumber of runs=3, Filtering=unfiltered2026.02 | 65.2 | 65.2 | 49.2 | 87 | 64.9 | — | — | |
| ORGANModel category=Sequence-Based2026.06 | 37.9 | 84.1 | — | — | 68.7 | — | — |