zengtsysu/BioNavi

Introduction

Source code for the paper, "Developing BioNavi for Hybrid Retrosynthesis Planning", published in JACS Au

Setup

BioNavi requires anaconda with Python 3.7 or later.

You will wish to install BioNavi in a virtual environment to prevent conflicting dependencies.

sudo apt-get install graphviz # optional: for output visualization
conda create -n bionavi python==3.7
conda activate bionavi
sh install.sh

Precursor prediction

We provide the pre-trained ensemble models for precursor prediction, one needs to first download the checkpoints and unzip under transformer/checkpoints/. You can also train your models with the following steps:

cd transformer

Preprocess the data

onmt_build_vocab -config config/vocab.yaml -n_sample -1

Train the model

onmt_train -config config/train.yaml

Note: Four models were trained with different seeds to obtain the ensemble model.

Make prediction

onmt_translate -src data/test_src.txt -output prediction/test_pred.txt \
                     -batch_size 32 -replace_unk -beam_size 10 -n_best 10 -gpu 0 -max_length 200 \
                     -model checkpoints/retro_seed_0.pt checkpoints/retro_seed_1024.pt \
                            checkpoints/retro_seed_1234.pt checkpoints/retro_seed_2023.pt

evaluate the prediction

python eval_single.py -targets data/test_tar.txt \
    -predictions prediction/test_pred.txt \
    -invalid_smiles -beam_size 10

Pathway planning

Now return to the main directory

cd ..

You can set the parameters by config/bionavi_conf.yaml, then run

python run_bionavi.py

Testing on the test set

Our work on BioNavi features an evaluation on our test set. We automated this evaluation by

python run_bionavi_batch.py

Evaluate the pathway prediction

python script/eval_pathway.py