Deep learning in GPCR drug discovery: Benchmarking the path to accurate peptide binding

Supporting data and source code for Deep learning in GPCR drug discovery: Benchmarking the path to accurate peptide binding. Summary Deep learning (DL) methods have drastically advanced structure-based drug discovery by predicting protein structures directly from sequences. However, predicting interactions between G-protein coupled receptors (GPCRs) and peptide ligands, remains underexplored. We evaluated DL tools including AlphaFold 2.3 (AF2), AlphaFold 3, RoseTTAFold-AllAtom, Peptriever, ESMFold, and D-SCRIPT for predicting GPCR-peptide interactions, benchmarking on both similar and dissimilar decoys. Structure-aware models performed best, with AF2 achieving the highest classification accuracy and how re-scoring on local interactions can further improve classification performance. When tested on 52 recent GPCR-peptide complexes, AF2 outperformed other methods, reproducing correct binding modes in nearly all cases. These results demonstrate that DL models can reliably re-discover peptide binders, aiding peptide drug discovery and guiding the selection of optimal tools for GPCR-targeted therapies. To this end, we provide a practical guide for selecting the best models for specific applications. The code is also publicly available on GitHub: https://github.com/HauserGroup/GPCR_peptide_benchmarking