Molecular datasets from "SMILES-Based Deep Generative Scaffold Decorator for De-Novo Drug Design"

Herein find the molecular datasets from "SMILES-Based Deep Generative Scaffold Decorator for De-Novo Drug Design". These were generated with SMILES-based scaffold decorator generative models trained with two training sets (DRD2 and ChEMBL). These generative models require a partially-built molecule (scaffold) as input and output several possible completions for each scaffold. Each dataset corresponds to a model trained with the  ChEMBL or DRD2 sets, wither multi-step (ms) or single-step (ss) and the provenance of the scaffolds (validation set, or non-dataset).

The molecules generated are annotated with a set of descriptors. The DRD2 datasets have the predicted probability of each molecule to be active on DRD2 (p) obtained from a Random Forest model. The ChEMBL model's descriptors are related to the synthesizability of the molecules (see manuscript). Also, the datasets decorated from validation set scaffolds are annotated whether they are part of the validation set (in_validation).