Overview

This repository contains the datasets used in "RetroSynFormer: Planning multi-step chemical synthesis routes via a Decision Transformer".  The repository contains the building blocks, templates and routes for the Standard, Small and Large datasets. In addition, predicted N1 and N5 routes mentioned in the publication is also attached.

Code is available at: https://github.com/emmaryd/retrosynformer/tree/main

File Description

Building Blocks:

standard_building_blocks.csv → set of building block included in the standard dataset

small_building_blocks.csv → set of building block included in the small dataset

large_building_blocks.csv → set of building block included in the large dataset

Reaction Templates:

standard_reaction_templates.pickle → reaction templates included in the standard dataset

small_reaction_templates.pickle → reaction templates included in the small dataset

large_reaction_templates.pickle → reaction templates included in the large dataset

Routes:

standard_routes.json → routes in the standard dataset

small_routes.json → routes in the small dataset

large_routes.json → routes in the large dataset

Predicted routes:

predicted_routes_bw50_n1.json → predictions for N1 test set, using the model trained on standard dataset, using beam width = 50

predicted_routes_bw50_n5.json → predictionsfor N5 test set, using the model trained on standard dataset, using beam width = 50