Release v2.0.0

This repository contains the train, validation, and test datasets used for training the FlowER model, along with the model checkpoint files. These checkpoint files allow full reproducibility of all data used in the FlowER paper.

There are three checkpoints for;

• FlowER model trained on dataset, which allows full reproducibility of all main Figures in the FlowER paper
• FlowER-large model trained on dataset, which allows a reproducibility of Figure 2a in the FlowER paper
• FlowER-large model trained on revisited dataset, which we have corrected minor errors in the mechanistic templates including intramolecular proton transfer reactions

There are two datasets;

• flower_dataset: the mechanstic dataset for full reproducibility of all main data used in the FlowER paper.
• flower_new_dataset: revisted mechanistic dataset which we have corrected minor errors in the mechanistic templates

In both datasets, the train, validation, and test datasets are recorded at the elementary step level. Each line in the text files follows the format:
reactants>>products|Number
Reactants and products are represented with atom-mapped SMILES.
The Number represents the overall reaction index. That is, elementary steps with the same Number originate from the same overall reaction.

Release v1.0.0

This repository contains the train, validation, and test datasets used for training the FlowER model, along with the model checkpoint files. These checkpoint files allow full reproducibility of all data used in the FlowER paper. (Dec 2024 training data)

The train, validation, and test datasets are recorded at the elementary step level. Each line in the text files follows the format:
reactants>>products|Number
Reactants and products are represented with atom-mapped SMILES.
The Number represents the overall reaction. That is, elementary steps with the same Number originate from the same overall reaction.