DeorphaNN: Virtual screening of GPCR peptide agonists using AlphaFold-predicted active-state complexes and deep learning embeddings

DeorphaNN DeorphaNN is a graph neural network that prioritizes peptide agonists for G protein-coupled receptors (GPCRs) by integrating active-state GPCR-peptide predicted structures, interatomic interactions, and deep learning protein representations. Dataset Details GPCR-peptide complexes from the nematode, C. elegans. Dataset Description Each of the 20035 GPCR-peptide complexes were run through AlphaFold-Multimer with active state templates. Active state predicted structures for dataset GPCRs are provided. Also provided are the relaxed predicted strucutres of the model with the highest peptide pLDDT per GPCR-peptide complex and the Arpeggio-identified intermolecular contacts, as well as regions of the pair representation averaged across all five models. The pair representation is processed to include a GPCR and peptide-specific region ({GPCR}_gpcr_T.hdf5 and {GPCR}_pep_T.hdf5) and the interaction region of the pair representation ({GPCR}_interaction.hdf5). Model weights for DeorphaNN are also provided, along with DeorphaNN scores for various orphan GPCRs. Dataset Sources Beets et al. Cell Reports, 2023 Code The DeorphaNN repo can be found at https://github.com/Zebreu/DeorphaNN