Data and Model for Quantifying the Failure Modes of Current One-step Retrosynthesis Models
Abstract
data_uspto/ contains: Clean & processed USPTO data (with original and RXNMapper-remapped reaction SMILES) complexity_features/: Computed complexity features 75_5_20_same_complex_doc_split/: Document-based split 75_5_20_random_split/: Random split 75_5_20_same_complex_doc_split_results/: Model evaluation and complementary metrics for further inspection Top-k accuracy summaries models_uspto/ contains the model checkpoints and vocabulary files for the Augmented Transformer (AT) and Graph2SMILES (G2S) models, as well as the model checkpoints and template files for the Template-Relevance (TR) model.