PAUL is a neural network based scoring function that is trained to classify models of protein-protein complexes.
The network is designed to classify structures generated with the docking program ATTRACT.
The architecture behind PAUL builds on rotation-equivariant tensor field networks.
The server expects each protein-protein model to contain two chains: A (receptor) and B (ligand).
To score a batch of candidate structures using PAUL, please upload a single PDB file below.
Please do not refresh the website until the file upload (as indicated by the progress bar) is complete.
The results will be emailed to the address specified upon job completion.
PDB files related to Eismann et al. (2020) can be downloaded here.
Each PDB file contains 1000 models generated with ATTRACT.