grail_metabolism 0.1.2

GRAIL is an open-source tool for drug metabolism prediction, based on graph neural networks and SMARTS reaction rules.

GRAIL: Graph neural networks and Rule-based Approach In drug metaboLism prediction

GRAIL is an open-source tool for drug metabolism prediction, based on SMARTS reaction rules and graph neural networks.

1. Installation

1.1 From source with Poetry

Run poetry install from the directory with pyproject.toml file

1.2 From PyPi

pip install grail_metabolism

IMPORTANT: If you are going to run GRAIL with CUDA, then after installation run install.py script to add proper versions of torch-geometric, torch-scatter and torch-sparse to your environment.

2. Data availability

Data can be downloaded from Zenodo draft. ATTENTION: This is not the final version of the dataset.

3. Quick start

IMPORTANT: Due to RXNMapper incompatibility with newer versions of Python, use only Python 3.9 or lower if you want to create your own set of transformation rules. All necessary tools are in grail.utils.reaction_mapper

For a quick start you may look into the notebooks/Unit_Tests.ipynb.

MolFrame

For the data uploading and further usage you should import grail_metabolism.utils.preparation.MolFrame. It has three different variants of initialization: from pandas.DataFrame, from dictionaries with metabolic maps, and from SDF file. For loading data from file use the MolFrame.from_file function, having previously read (substrate, metabolite, real_or_not) triples via MolFrame.read_triples.

Models

In the model module you can find all necessary model classes, especially Filter and Generator.