PeptideGNN

A GNN-RNN hybrid model for predicting and explaining peptide LC-RT data.

Overview

To reproduce the results of the paper, check out Run it yourself.

To build your own GNN-RNN hybrid model on your own datasets, check out Build it yourself.

Run it yourself

Note

Results may differ from the paper due to the stochastic nature of neural nets.

1. Clone the repository

git clone [email protected]:CompOmics/peptide-gnn.git
cd peptide-gnn

2. Installation

We recommend using a virtual environment. Choose the installation that matches your hardware:

For GPU (CUDA) users:

pip install .[gpu]

For CPU users:

pip install .

3. Train and explain

Run the training pipeline using the provided sample data. By default, results are saved to ./output/.

pepgnn run ./data/ --epochs 100

4. Visualize results

We provide a Jupyter notebook for post-hoc analysis and visualization:

1. Launch Jupyter: jupyter notebook
2. Navigate to notebooks/vis.ipynb
3. Run all cells to generate plots and model explanations.

Build it yourself

If you are interested in building your own GNN-RNN hybrid model, check out MolCraft.

Note

While MolCraft provides an improved API for molecular GNN-building, the underlying building blocks differ slightly from MolGraph, which may result in different outcomes.