Table 3 Comparative performance of three ChemBERTa-based models on the twelve QM9 properties. Each row corresponds to a specific QM9 property, while the columns list R² and RMSE for (i) ChemBERTa-zinc-base-v1 (“v1”), (ii) ChemBERTa-77 M-MLM (“v2”), and (iii) the SELFIES-domain-adapted model (“SELFIES DA”). The SELFIES-domain-adapted model outperforms the smaller SMILES model in all targets and in most cases exceeds the larger model’s performance.
From: Domain adaptation of a SMILES chemical transformer to SELFIES with limited computational resources
Property | R² | RMSE | ||||
|---|---|---|---|---|---|---|
v1 | v2 | SELFIES DA | v1 | v2 | SELFIES DA | |
μ | 0.6725 ± 0.0078 | 0.7439 ± 0.0054 | 0.7440 ± 0.0067 | 0.8593 ± 0.0124 | 0.7598 ± 0.0048 | 0.7596 ± 0.0083 |
α | 0.6809 ± 0.0075 | 0.9451 ± 0.0036 | 0.9686 ± 0.0022 | 4.6133 ± 0.0727 | 1.9123 ± 0.0609 | 1.4463 ± 0.0490 |
\(\in _{{HOMO}}\) | 0.7769 ± 0.0113 | 0.8874 ± 0.0022 | 0.8804 ± 0.0053 | 0.0104 ± 0.0003 | 0.0074 ± 0.0001 | 0.0076 ± 0.0002 |
\(\in _{{LUMO}}\) | 0.9242 ± 0.0070 | 0.9578 ± 0.0021 | 0.9526 ± 0.0040 | 0.0129 ± 0.0006 | 0.0096 ± 0.0002 | 0.0102 ± 0.0004 |
\(\in _{{gap}}\) | 0.8959 ± 0.0054 | 0.9381 ± 0.0013 | 0.9352 ± 0.0014 | 0.0152 ± 0.0004 | 0.0117 ± 0.0001 | 0.0120 ± 0.0001 |
< R²> | 0.6552 ± 0.0124 | 0.9341 ± 0.0027 | 0.9514 ± 0.0037 | 164.7080 ± 3.7830 | 72.0325 ± 2.1329 | 61.7844 ± 2.3331 |
zpve | 0.8578 ± 0.0082 | 0.9677 ± 0.0024 | 0.9813 ± 0.0016 | 0.0125 ± 0.0003 | 0.0059 ± 0.0002 | 0.0045 ± 0.0002 |
U0 | 0.7245 ± 0.0129 | 0.9644 ± 0.0011 | 0.9828 ± 0.0012 | 20.9064 ± 0.5646 | 7.5164 ± 0.0979 | 5.2261 ± 0.1893 |
U | 0.7436 ± 0.0063 | 0.9653 ± 0.0011 | 0.9832 ± 0.0008 | 20.1722 ± 0.3222 | 7.4171 ± 0.1182 | 5.1588 ± 0.1020 |
H | 0.7420 ± 0.0088 | 0.9637 ± 0.0021 | 0.9826 ± 0.0012 | 20.2327 ± 0.3604 | 7.5918 ± 0.2141 | 5.2573 ± 0.1938 |
G | 0.7468 ± 0.0049 | 0.9657 ± 0.0034 | 0.9819 ± 0.0014 | 20.0472 ± 0.2422 | 7.3691 ± 0.3614 | 5.3600 ± 0.1806 |
Cv | 0.7769 ± 0.0069 | 0.9591 ± 0.0019 | 0.9729 ± 0.0014 | 1.9192 ± 0.0239 | 0.8222 ± 0.0232 | 0.6689 ± 0.0156 |
