GEMS – Enhancing Generalizable Binding Affinity Prediction by Removing Data Leakage and Integrating Language Model Embeddings into Graph Neural Networks

The field of computational drug design requires accurate scoring functions to predict binding affinities for protein-ligand interactions. However, train-test data leakage between the PDBbind database and the CASF benchmark datasets has significantly inflated the performance metrics of currently available deep-learning-based binding affinity prediction models, leading to overestimation of their generalization capabilities. We address this issue by proposing PDBbind CleanSplit, a training dataset curated by a novel structure-based filtering algorithm that eliminates train-test data leakage as well as redundancies within the training set. Retraining current top-performing models on CleanSplit caused their benchmark performance to drop significantly, indicating that the performance of existing models is largely driven by data leakage. In contrast, our graph neural network model for efficient molecular scoring (GEMS) maintains high benchmark performance when trained on CleanSplit. Leveraging a sparse graph modeling of protein-ligand interactions and transfer learning from language models, GEMS is able to generalize to strictly independent test datasets.