Table 2 Description of all molecular featurizers benchmarks
From: Learning self-supervised molecular representations for drug–drug interaction prediction
Featurizer | Type | Dataset name | Dataset size | Feature vector. dim | Architecture |
|---|---|---|---|---|---|
ECFP | Hashed fingerprint | – | – | 2048 | – |
ChemBERTa-77M | Pretrained | PubChem | 77 M | 384 | Transformer |
MOL2VEC | Pretrained | ZINC + ChemBL | 19.9 M | 300 | Word2vec |
SMR-DDI | Pretrained | Chembl | 200 K | 262 | CNN |
ChemGPT-1B | Pretrained | PubChem | 10 M | 256 | Transformer |
MACCKEYS | Structural fingerprint | – | – | 166 | – |
gin_supervised_edgepred | Pretrained | ChembL + ZINC15 | 465 K + 2 M | 300 | Graph |
ChemGPT-4M | Pretrained | PubChem | 10 M | 128 | Transformer |
gin_supervised_contextpred | Pretrained | ChembL + ZINC15 | 465 K + 2 M | 300 | Graph |
gin_supervised_masking | Pretrained | ChembL + ZINC15 | 465 K + 2 M | 300 | Graph |
gin_supervised_masking | Pretrained | ChembL + ZINC15 | 465 K + 2 M | 300 | Graph |