GraphGPSM is an equivariant graph neural network based scoring model that can score monomeric and complex protein structures. According to the protein structure provided by the user, GraphGPSM will automatically calculate the residue-level USR, the distance and direction encoded by Gaussian radial basis function, the Rosetta energy term, the backbone dihedral angle, the distance and orientations between residues to characterize the protein structure, and then these features are fed into the trained graph neural network to calculate the score. GraphGPSM's scores are highly correlated with the model's true TM-score. GraphGPSM can be used to guide protein folding, which can further improve the structure predicted by AlphaFold2. GraphGPSM (group name: GuijunLab-Threader) has achieved competitive performance in the SCORE ranking of CASP15 evaluation model accuracy prediction.

• 2022/12/13: GraphGPSM (Group name: GuijunLab-Threader) has achieved a high ranking in SCORE ranking in CASP15 evaluation model accuracy prediction.
• 2022/05/01: GraphGPSM development is complete.

• Guangxing He^✝,Jun Liu^✝,Dong Liu and Guijun Zhang*. GraphGPSM: A Global Scoring Model for Protein Structure Using Graph Neural Networks. Submitted.