DFRscore

Code for the paper: "DFRscore: Deep Learning-Based Scoring of Synthetic Complexity with Drug-Focused Retrosynthetic Analysis for High-Throughput Virtual Screening" by Hyeongwoo Kim, Kyunghoon Lee, Chansu Kim, Jaechang Lim, and Woo Youn Kim.

Table of Contents

• Install Dependencies
• Preparing Training Data
• Retrosynthetic analysis
• Training DFRscore
• Using DFRscore to Evaluate Synthetic Accessibility

Install Dependencies

DFRscore needs conda environment. After installing conda,
you can manually install the required pakages as follows:

• rdkit=2020.09.1
• matplotlib
• scipy
• numpy
• scikit-learn
• pytorch>=1.11.0

Or simply you can install the required packages by running

. ./dependencies

This will configure a new conda environment named 'DFRscore'.

Preparing Training Data

1. If you ONLY want to train the DFRscore model without retrosynthetic analysis, you can get the data for training with:

./unzip_train_data.sh

And go to Training DFRscore.

2. Or, if you want to get the training data by running our retrosynthetic anaylsis tool (FRA) followed by model training, you should download the ingredients and put them at data/retro_target_data. Every line of the file should inclue a single molecule of SMILES representation. You can filter them with data/retro_target_data/molecules_filter.py.

And go to Retrosynthetic Analysis.

Retrosynthetic Analysis

After preparing the target molecule data for retrosynthetic analysis, you can run FRA by:

python retro_analysis.py \
--template data/template/retro_template.pkl \
--reactant data/reactant_bag/filtered_mcule.smi \
--retro_target <PATH-TO-TARGET-FILE> \
--depth 4 \
--num_molecules <NUMBER-OF-MOLECULES-TO-BE-ANALYZED> \
--num_cores <NUMBER-OF-CPU-CORES>

And go to Training DFRscore.

Training DFRscore

After getting data for training by ./unzip_train_data.sh or manually running ./retro_analysis.py, you can train a new model by:

python ./train_model.py

Using DFRscore to Evaluate Synthetic Accessibility

Useful functions are described in DFRscore class in scripts/modelScripts/model.py file.

For a fast test, you can simply run python getDFRscore.py --model_path <path_to_model_dir> --smiles <SMILES>

For example, python get_DFRscore.py --model_path ./save/PubChem/DFRscore/Best_model_163.pt --smiles "CC(=O)Nc1ccc(cc1)OCC"