[v1.0] (https://github.com/camlab-ethz/GEMS) (2025-07-15) Initial release of 'GNN for Efficient Molecular Scoring, GEMS, a graph-based deep learning model designed for protein-ligand binding affinity prediction. It includes instructions for installing dependencies, preparing datasets, training the model, and running inference. The repository also features PDBbind CleanSplit, a refined training dataset based on PDBbind that minimizes data leakage and enhances model generalization. Features Graph neural network for binding affinity prediction Filtering Algorithm that created PDBbind CleanSplit Search algorithm for detecting data leakage in protein-ligand structural datasets