IDEA-AI4S/MolInst_RS_125K_Scaffold_SMILES-MMChat

Retrosynthesis Prediction Dataset (derived from MolInstruct) molecule representation format: 1D SMILES will further encode into 2D graph features We use scaffold splitting to reconstruct the train-split. We use SMolInstruct RS train split as the sample pool. We only include single-step retrosynthesis prediction. For Detail, refer to PRESTO: Progressive Pretraining Enhances Synthetic Chemistry Outcomes: https://arxiv.org/pdf/2406.13193