rxn-neb

Setup

To install the package run:

poetry install

Pre-process data for running rxn-neb

Here we assume to start from a JSON file reporting synthesis trees characterized by reactions SMILES represented using a pre-order traversal. A sample is provided here.

Additionally we assume a model for reaction fingerprints compatible with rxnfp is available (see the repo for instructions on how to train your own on public or proprietary data).

To get the default model used in RXN for Chemistry simply clone the repo:

git clone https://github.com/rxn4chemistry/rxnfp.git

You can directly use the default model available at ./rxnfp/rxnfp/models/transformers/bert_ft.

Prepare the fingerprints from available synthesis trees:

generate-fingerprints --reaction_trees_path "./sample-data/reaction_trees.json" --fingerprints_model_path "./rxnfp/rxnfp/models/transformers/bert_ft" --generated_fingerprints_path "./sandbox/generated_fingerprints.csv"

Prepare the PCA model for fingerprint compression and related indexes:

generate-pca-compression-and-indices  --reaction_trees_path "./sample-data/reaction_trees.json" --fingerprints_path "./sandbox/generated_fingerprints.csv" --pca_model_filename "./sandbox/pca.pkl" --tree_data_dict_pca_filename "./sandbox/tree_data_dict_pca.pkl"

NOTE: these examples are creating a sandbox folder where all outputs are stored.

Usage

We assume you have a pair of single-step forward and backward model trained using rxn-onmt-models (see the repo for a detailed guide on how to train them on public or proprietary data).

run-neb-retrosynthesis --product "NS(=O)(=O)c1nn(-c2ccccn2)cc1Br" \
    --forward_model_path "/path/to/forward_model.pt" \
    --backward_model_path "/path/to/backward_model.pt" \
    --fingerprints_model_path "./rxnfp/rxnfp/models/transformers/bert_ft"  \
    --pca_model_filename "./sandbox/pca.pkl" \
    --tree_data_dict_pca_filename "./sandbox/tree_data_dict_pca.pkl" \
    --output_path ./test_retro.json