topological-pretraining

This repository contains the code for "On Improving Graph Neural Networks for QSAR via pre-training on Extended-Connectivity Fingerprints". Preprint can be found here.

Installing within conda env

git clone git@github.com:oxpig/topological-pretraining.git
cd topological-pretraining
conda env create --file environment.yaml
conda activate topological-pretraining

For using CUDA, change cpu in https://data.pyg.org/whl/torch-2.8.0+cpu.html to cu{CUDA VERSION} (e.g., cu124) in the environment.yaml.

Notebooks

See:

• Quickstart notebook for basic usage.
• Dataset notebook for loading and setting up datasets.
• Models notebook for basic model demonstrations.
• Benchmark evaluations for statistical tests and visualisations.
• Substructure importance notebook for estimating pre-trained GIN substructure importance.