Table 6 Performance comparison of GMT models with different representations
From: Probabilistic generative transformer language models for generative design of molecules
GMT models | ||||||||
|---|---|---|---|---|---|---|---|---|
GMT- SMILES | GMT-QM9- SMILES | GMT-PE- SMILES | GMT- SELFIES | GMT- DEEP | GMT-PE- DEEP | |||
Validity | \(\uparrow\) | 0.8586 | 0.8937 | 0.8288 | 1.0000 | 0.8168 | 0.7954 | |
Unique@1k | \(\uparrow\) | 1.0000 | 1.0000 | 1.0000 | 1.0000 | 1.0000 | 1.0000 | |
Unique@10k | \(\uparrow\) | 0.9998 | 0.9689 | 0.9995 | 1.0000 | 1.0000 | 0.9997 | |
IntDiv | \(\uparrow\) | 0.8569 | 0.9182 | 0.8558 | 0.8701 | 0.8570 | 0.8519 | |
Filters | \(\uparrow\) | 0.9765 | 0.6549 | 0.9797 | 0.7961 | 0.9844 | 0.9847 | |
Novelty | \(\uparrow\) | 0.9532 | 1.0000 | 0.8829 | 0.9683 | 0.9367 | 0.9149 | |
Test | 0.5381 | 0.2575 | 0.5778 | 0.4673 | 0.5509 | 0.5722 | ||
SNN | \(\uparrow\) | TestSF | 0.5143 | 0.2510 | 0.5460 | 0.4485 | 0.5246 | 0.5405 |
Test | 0.7294 | 30.5280 | 0.1986 | 3.7750 | 0.3604 | 0.4366 | ||
FCD | \(\downarrow\) | TestSF | 1.2607 | 31.3022 | 0.7595 | 4.5698 | 0.9563 | 1.0736 |
Test | 0.9879 | 0.3945 | 0.9982 | 0.9869 | 0.9981 | 0.9967 | ||
Frag | \(\uparrow\) | TestSF | 0.9850 | 0.3909 | 0.9958 | 0.9831 | 0.9964 | 0.9934 |
Test | 0.8661 | 0.0007 | 0.9125 | 0.8431 | 0.8880 | 0.8903 | ||
Scaf | \(\uparrow\) | TestSF | 0.1649 | 0.0000 | 0.1087 | 0.1096 | 0.1511 | 0.1170 |