ADME (Absorption, Distribution, Metabolism and Excretion) prediction for small molecules using three different models:

1. A graph neural network
2. A random forest
3. A transformer

Results are computed and logged in the results.txt file.

To build the environment, run the following commands in a shell:

$ python -m venv env
$ source env/bin/activate
$ (env) source setup.sh

Note that only some of the dependencies are in the requirements.txt file; merely installing these is not sufficient. The reason the setup.sh script exists is because PyTorch Geometric needs to be installed from source and involves several steps which need to be executed in sequence, and one of its dependencies needs to be installed directly from GitHub as it not hosted on PyPi.

To create a Jupyter kernel, run:

$ (env) ipython kernel install --name "env" --user

Note: The setup instructions above have only been tested for python 3.10 on M1 Mac.

The Jupyter notebook has been converted to markdown using Jupytext (to reduce storage requirements on GitHub). To convert it back to .ipynb format and run it, run the following commands:

$ (env) jupytext --to ipynb *.md 
$ (env) jupyter notebook

The dataset used is the CYP P450 2C19 Inhibition dataset from Therapeutic Data Commons. The task is binary classification of CYP2C19 inhibition given a drug's SMILE string.