Heterogeneous Conditional Molecular Generation on 10K Molecules Drug-related task set
100ValidityGraph GA
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Graph GA2026.05 | 100 | 88.06 | 96.58 | 7.6709 | 2.0405 | 46.91 | 30.17 | 12.3333 | |
| MARS2026.05 | 100 | 85.71 | 79.14 | 8.9636 | 2.3988 | 53.2 | 53.89 | 9 | |
| JTVAE-BO2026.05 | 100 | 69.96 | 60.78 | 33.4123 | 1.9124 | 46.32 | 49.55 | 11 | |
| FREED2026.05 | 99.8 | 83.26 | 10.18 | 20.609 | 2.4163 | 77.56 | 92.13 | 7.3333 | |
| LSTM-HC2026.05 | 98.02 | 91.22 | 85.9 | 14.8023 | 1.8267 | 50.76 | 52.42 | 8 | |
| CoMole w/o RLAlignment=Supervised Fine-Tuning2026.05 | 96.68 | 88.16 | 96.16 | 7.6326 | 1.3169 | 95.04 | 94.39 | 2.3333 | |
| CoMoleAlignment=Reinforcement Learning2026.05 | 95.7 | 84.22 | 64.01 | 17.6755 | 1.0764 | 100 | 95.5 | 1 | |
| Graph DiT2026.05 | 83.44 | 87.32 | 91.47 | 9.113 | 1.239 | 91.33 | 93.42 | 2.6667 | |
| GDSS2026.05 | 79.28 | 89.68 | 57.33 | 9.5036 | 3.577 | 50.36 | 50.44 | 12 | |
| MOOD2026.05 | 76.19 | 91.26 | 19.01 | 37.2665 | 2.0941 | 50.66 | 48.21 | 11.3333 | |
| GDPO2026.05 | 73.07 | 89.92 | 71.26 | 5.3391 | 1.3598 | 84.57 | 85.56 | 4.3333 | |
| DiGress v22026.05 | 51.72 | 90.16 | 71.6 | 7.5507 | 1.8906 | 52.15 | 67.71 | 7 | |
| DiGress2026.05 | 50.61 | 89.61 | 69.59 | 6.5017 | 1.9023 | 52.63 | 67.26 | 7.3333 | |
| DeFoG2026.05 | 50.34 | 92.65 | 38.36 | 8.1281 | 1.9216 | 55.87 | 46.36 | 9.3333 |