Use this page to diagnose the problems you hit while running P2Rank — PDB parsing errors, no pockets detected, a top pocket that misses the known site, weak predictions on ESMFold/AlphaFold or low-resolution structures, too many pockets, and PyMOL rendering issues. Preparing the input structure correctly and picking the right model variant resolves most of them.

Common issues and fixes

P2Rank uses random forest machine learning to predict binding pockets from protein 3D structure. Understanding common issues helps you prepare inputs correctly and interpret predictions appropriately.

PDB parsing errors

Ensure the PDB file contains valid ATOM records with correct formatting (columns must be properly aligned)
Check that chain identifiers are present and consistent throughout the file
Remove HETATM records for non-standard residues if they cause parsing failures (P2Rank ignores them anyway)
For multi-model PDB files (NMR ensembles), only the first MODEL record is used; ensure it is complete
Verify that the PDB ID is correct and exists in RCSB PDB; typos in accession codes are a common error

No pockets detected

Very small proteins (<50 residues) may lack sufficient surface area for pocket detection; consider whether the protein realistically has binding pockets
Check that the submitted structure isn't just a small fragment — there's no chain-filter parameter, so if you need to analyze one chain, submit pdb_content containing only that chain
For predicted structures with large low-confidence regions, try the alphafold model variant which handles prediction uncertainty
Verify that the protein is properly folded in the structure; unfolded or disordered structures do not have pockets

Top pocket does not match known binding site

P2Rank ranks pockets by probability score, not by biological importance. The known active site may rank second or third
Check whether the known binding site is occupied by a ligand in the PDB; P2Rank removes ligands before prediction, and the apo pocket may score differently
For multi-domain proteins, the top-scoring pocket may be on a different domain than expected; there's no chain-filter parameter, so submit a pdb_content trimmed to the domain of interest if you need to focus the analysis

Poor predictions on ESMFold/AlphaFold structures

Always use the alphafold model variant for predicted structures; the default model is calibrated on experimental structures
Low-confidence regions (pLDDT < 50) in predicted structures may create artificial pockets at the protein surface; check pLDDT scores for pocket-lining residues
Predicted structures may have subtle geometric differences from true structures that affect pocket detection
Consider trimming low-confidence tails or loops before running P2Rank to reduce false positives

Low-resolution structure issues

Structures with resolution >3.0 angstroms may produce less accurate pocket predictions due to imprecise atomic coordinates
Missing side chains in low-resolution structures reduce the accuracy of surface feature computation
Consider rebuilding missing side chains with homology modeling or Rosetta before running P2Rank
Use multiple structures of the same protein (if available) and compare pocket predictions for consensus

Too many pockets detected

Large multi-chain complexes naturally have many surface concavities; use the max_pockets parameter to limit output (1-50, default 10)
There's no minimum-score threshold parameter — filter the returned pocket table by pocket_score yourself, or focus on pockets with druggability classified as "druggable" or "intermediate"
For multi-chain structures, there's no chain-filter parameter — submit pdb_content trimmed to the chain(s) of interest for a focused analysis

PyMOL script does not render correctly

Ensure PyMOL is installed and the PDB file is accessible at the path referenced in the script
The PyMOL script assumes the original PDB file is in the same directory; adjust paths if needed
Color scheme may not display correctly in PyMOL versions prior to 2.0; update to the latest version
If pockets overlap with crystallographic contacts, the visualization may be cluttered; hide symmetry mates in PyMOL

Still stuck?

Provide the job ID, PDB ID or filename, model variant, and any error messages
Note the number of chains, total residue count, and structure source (experimental vs. predicted)
Share whether the issue is with structure parsing, pocket detection, or result interpretation
Include the known binding site residues (if available) for comparison with predictions

P2Rank troubleshooting