Table 4 RMSE on ESOL, FreeSolv, and lipophilicity. SELFIES FT refers to the domain-adapted model fine-tuned end-to-end on each task. Despite limited pretraining scale and constrained computational resources, it performs competitively across all benchmarks.
From: Domain adaptation of a SMILES chemical transformer to SELFIES with limited computational resources
Model | RMSE | ||
---|---|---|---|
ESOL | FreeSolv | Lipophilicity | |
SELFIES FT | 0.944 | 2.511 | 0.746 |
SELFormer30 | 0.682 | 2.797 | 0.735 |
ChemBERTa-77M20 | 1.025 | - | 0.987 |
D-MPNN23 | 1.050 | 2.082 | 0.683 |
AttentiveFP68 | 0.877 | 2.073 | 0.721 |
N-GramRF67 | 1.074 | 2.688 | 0.812 |
N-GramXGB67 | 1.083 | 5.061 | 2.072 |
PretrainGNN66 | 1.100 | 2.764 | 0.739 |
GROVERbase18 | 0.983 | 2.176 | 0.817 |
GROVERlarge18 | 0.895 | 2.272 | 0.823 |
GEM7 | 0.798 | 1.877 | 0.660 |