synformer 0.1.0

Synformer: Generative Model for Synthesizable Molecule Generation

synformer

This repo contaisn the main code for SynFormer model. To run the model, you need to first download the list of building blocks. We only provide a 5k building block list for test purpose.

TO-DOs

• Clean the repo and prepare for the release
• Make a demo on Huggingface

Instruction

0. Start (for development)

pip install -e .

1. Preprocess data

Download building block data to data/building_blocks/ in sdf format .

python scripts/preprocess.py --model-config configs/dev_smiles_diffusion.yml

2. Train the model

python scripts/train.py configs/dev_smiles_diffusion.yml --debug --devices 1
# python scripts/train_ed.py configs/dev_ed.yml --debug --devices 1

3. Model inference

Work in progress

4. Use for molecular optimization

python scripts/molopt.py \
    --model <decoder-only-model-checkpoint> \
    --use-replay-buffer \
    --use-prior \
    --oracle <domain>:<task>  # Example: "tdc:osimertinib_mpo"