Using pretrained GROVER to extract the atomic fingerprints from molecule. The fingerprints can be used for further tasks.

GROVER is short for Graph Representation frOm self-superVised mEssage passing tRansformer which is a Transformer-based self-supervised message-passing neural network by Rong and colleagues as in the paper: Self-Supervised Graph Transformer on Large-Scale Molecular Data.

1. Create and activate a conda environment:

conda create --name grover python=3.6.8
conda activate grover

2. Install requirements from requirements.txt file. Additionally, install torchinfo:

conda install -c conda-forge -c pytorch -c acellera -c RMG --file=requirements.txt
pip install torchinfo

There are two pretrained models provided by the original authors. Download, extract and save the .pt file in models_pretrained/.

• GROVER_base
• GROVER_large

Run the main.py file:

python main.py

Details about the arguments can be viewed in the setup_parser() function found in the main.py, or by running:

python main.py -h

If no arguments are specified, then the default arguments will be used.

By default, the outputs are saved in extracted_fingerprint. The outputs include 3 files:

• atom_fp.npy: contains the atomic fingerprints.
• distance.npy: contains the pair-wise shortest relative distance matrices between nodes of the molecular graphs.
• smiles.txt: contains the SMILES strings of the molecules.

In order to read the .npy files, please refer to this part in the numpy.save documentation