While working with the service, one should keep in mind that:
  • PISA is not a curated source of data, but rather presents results calculated by certain physico-chemical models for PDB and/or uploaded macromolecular structures.  
  • Estimates for macromolecular-ligand interactions in PISA are less precise than those for macromolecular-macromolecular interactions. Ligands in PISA are effectively considered as surface modifiers (fixed to the macromolecular surface), unless that is not possible due to symmetry of interactions. Fixed ligands are not included into the dissociation pattern. 
  • PISA relies on the correctness of coordinate files submitted. "Correctness" implies not only a formal compliance with the file format, but also using standard names for atoms and residues. Using trivial names prevents PISA of getting data from the database, which will result in ignoring these unrecognised residues. 
  • Incomplete coordinate files (such as backbone-only) are not suitable for PISA unless only a few residues from many are affected.  
  • PISA uses identical computational procedures for PDB entries and user uploads.  
  • The physical-chemical models, used in PISA, are implemented with a finite precision in the mathematical sense. As a result, some figures may not add up precisely. For example, the calculation of surface area produces slightly different results at different orientations of the molecule.  
  • It is stated, however, that to the best knowledge and belief of the authors, the finite precision of implementation does not result in artifacts beyond those inherent in the used models.  
  • The models, used in PISA, have their validity limits. Therefore the results, such as macromolecular oligomeric states, should be rated as probable. It is estimated that oligomeric states are correct in 85-90% of instances, however, there is no hard evidence to prove or disprove this figure.  
  • PISA models include calibration parameters that are tuned in order to achieve a higher probability of correct answers. From time to time these parameters may change, which will cause a change in the resulting figures. It is expected, however, that this change will not be significant in most instances.  
  • Interface structural searches are based on PDBeFold technology, which employs certain similarity thresholds. It is therefore possible that some lower-similarity matches, which may still be of interest, are missed by the service.