PDBePISA Tutorial: Analyse a Protein-Protein Interface
To analyse a protein-protein interface in PDBePISA, open the free EBI server, enter a PDB code or upload your complex as a PDB or mmCIF file with each partner on its own chain, and click Interfaces. Read the interface area, the solvation free energy gain (ΔiG), its P-value, the hydrogen bond and salt bridge counts and the CSS score, then open Details for the residue list.
PDBePISA (Proteins, Interfaces, Structures and Assemblies) is usually the first tool reviewers expect to see after a docking run, and it is the follow-on step to our guides on protein-protein docking with ClusPro, HDOCK and HADDOCK and docking a vaccine construct with TLR4. This tutorial uses two real PDB entries with the numbers PISA returns for them, so you can check your own reading against ours. It was written by the StemSkills Lab team, who have spent more than ten years on structural bioinformatics, drug design and molecular modeling.
What is PDBePISA and what does it calculate?
PISA is a method from Evgeny Krissinel and Kim Henrick at the European Bioinformatics Institute, published as “Inference of macromolecular assemblies from crystalline state” (J. Mol. Biol. 372, 774 to 797, 2007). Crossref records more than 10,000 citations of that paper, which is why reviewers recognise its output on sight. The PDBePISA service can:
- retrieve precalculated results for every entry in the PDB archive;
- calculate results for structures you upload as PDB or mmCIF files;
- report the surface and interface properties of each chain pair, the probable quaternary structures (assemblies) and how they would dissociate;
- search the PDB for similar interfaces formed by structural homologs.
For a student, three outputs matter most: the Interfaces table (one row per contact between two chains), the Interface Details page (which residues and which bonds), and the Assemblies table (which complex PISA thinks is stable in solution).
One practical note. The legacy PDBePISA page now carries a banner saying a new version of PISA is available and the legacy page will be retired in the future. The calculations and column names below come from the legacy service (it reports itself as PISA v1.52), which is still the version most published methods sections cite. If you switch to the new interface, record which version you used.
How do you run PDBePISA on a PDB entry or your own docked complex?
- Prepare the file (docked models only). Put each partner on its own chain ID (receptor A, ligand protein B), delete waters, and keep one model per file. Details are in the troubleshooting section.
- Open the service. Go to ebi.ac.uk/pdbe/pisa and click Launch PDBePisa.
- Choose the input. For a crystal structure, type the four-character PDB code. For your own complex, choose to upload a coordinate file and select your PDB or mmCIF file.
- Click Interfaces. PISA lists every unique interface it finds, sorted by interface area.
- Open the top interface. Select its row and click Details to see the interfacing residues, their accessible and buried areas, and the list of hydrogen bonds and salt bridges with distances.
- Check Assemblies. This tells you whether PISA predicts the complex to be stable in solution, with its buried area, ΔGint and ΔGdiss.
For any PDB entry you can also jump straight to a table with a link. The PISA linking page documents three tokens: qi (interface table), qs (structure table) and qa (assembly table). For example:
https://www.ebi.ac.uk/pdbe/pisa/cgi-bin/piserver?qi=1hsg
https://www.ebi.ac.uk/pdbe/pisa/cgi-bin/piserver?qa=1hsgUploads cannot be launched this way; they need the web form.
How do you read the PDBePISA Interfaces table?
Every row is one interface between “Structure 1” and “Structure 2”. The definitions below are taken from the help pages behind each column header in PISA itself.
| Column | What it means | How to use it |
|---|---|---|
| iNat, iNres | Number of interfacing atoms and residues on each side | Report the residue counts per partner |
| Surface Å2 | Total solvent accessible surface of each structure | Context only |
| Interface area, Å2 | Difference in accessible surface of the isolated and interfacing structures, divided by two | This is area per side, not total buried area (see below) |
| ΔiG, kcal/mol | Solvation free energy gain on forming the interface; negative means hydrophobic | Excludes hydrogen bonds and salt bridges, so it is not a binding energy |
| ΔiG P-value | Probability of a lower ΔiG if the same area were picked at random from the surface | Below 0.5 suggests a specific, unusually hydrophobic patch |
| NHB, NSB, NDS | Potential hydrogen bonds, salt bridges and disulphide bonds across the interface | Report them, and list the key pairs from Details |
| CSS | Complexation Significance Score, 0 to 1: how much the interface contributes to the stable assemblies PISA finds | Close to 1 means the interface holds a predicted stable assembly together |
Two of these are misread constantly.
The P-value is about hydrophobicity, not statistical significance of binding. PISA’s own help page says P = 0.5 means the interface is not surprising at all, and that P above 0.5 means the interface “is likely to be an artefact of crystal packing”. A low P-value says the patch is more hydrophobic than a random patch of the same size. It says nothing about polar or charged interfaces, which matters a lot in the barnase example below.
Interface area is half the buried surface. PISA computes it as (A + B − AB) / 2, so a value of 800 Å2 means roughly 1,600 Å2 of total buried surface area (BSA). Many papers quote BSA. Say which one you report, or you will appear to disagree with the literature by a factor of two.
What counts as a hydrogen bond or salt bridge in PISA?
The counts are “potential” bonds identified from geometry in the static structure. The Details page lists each pair with its atoms and distance. PISA’s help gives rough weights (about 0.5 kcal/mol per hydrogen bond and 0.3 kcal/mol per salt bridge in its own energy model) and notes that these depend on calibration and may change between versions. Treat the counts as descriptive, and check that the pairs make chemical sense before you write about them.
Want the guided, hands-on version?
Our live Molecular Modeling & MD Simulations cohort bootcamp takes you from zero to running real docking and MD workflows, with a portfolio project for your grad-school applications.
What does a real PDBePISA result look like? Two worked examples
These are the precalculated PISA results for two well-known entries, as the server returns them today.
HIV-1 protease dimer (PDB 1HSG): a clear biological interface
1HSG is HIV-1 protease bound to an inhibitor (ligand code MK1). Its top interface, chain A with chain B, gives:
- interface area 1,746.9 Å2, with 42 interfacing residues on each chain;
- ΔiG −22.5 kcal/mol, P-value 0.235;
- 32 hydrogen bonds and 7 salt bridges;
- CSS 0.974.
The Assemblies table agrees: PISA reports the dimer AB with the inhibitor as stable in solution, with 3,560 Å2 buried, ΔGint −23.5 kcal/mol and ΔGdiss 26.3 kcal/mol. A positive ΔGdiss is what PISA requires to call an assembly stable. Every signal points the same way, which matches the biology: the protease is an obligate dimer.
The inhibitor also appears as its own rows ([MK1]B:902 against chain A and chain B, about 390 Å2 each). PISA treats ligands as separate structures, so delete them if you only want the protein-protein interface.
Barnase with barstar (PDB 1BRS): a tight complex PISA undervalues
1BRS is the 2.0 Å crystal structure of barnase (chains A, B, C) bound to its inhibitor barstar (chains D, E, F), both from Bacillus amyloliquefaciens. Interface A with D gives:
- interface area 778.5 Å2;
- ΔiG +0.7 kcal/mol, P-value 0.591;
- 14 hydrogen bonds and 12 salt bridges;
- CSS 0.000.
The Assemblies page concludes that the chains most probably do not form a complex in solution. Yet barnase and barstar form one of the tightest protein complexes measured. Schreiber and Fersht (Biochemistry, 1993) report a dissociation constant of 1.3 × 10−14 M, which corresponds to a ΔG of −18.9 kcal/mol at 25 °C.
The reason is in the column definitions. This interface is polar and charged (12 salt bridges in 778.5 Å2), and ΔiG is a solvation term that ignores hydrogen bonds and salt bridges. PISA was built to separate biological assemblies from crystal contacts, and a small, electrostatic heterocomplex interface looks like a crystal contact to that model. This is the most useful lesson in this tutorial: PISA is evidence, not a verdict. Combine it with experimental data, conservation (EPPIC) or an affinity predictor (PRODIGY) before you claim a complex is or is not real.
How do you tell a biological interface from a crystal contact?
For crystal structures, use several signals together rather than one cut-off:
- CSS near 1 and the interface appearing in a stable assembly with positive ΔGdiss;
- P-value below 0.5 together with a negative ΔiG;
- a large area relative to the other interfaces in the same crystal (in 1HSG the next protein contact is 254.5 Å2, against 1,746.9 Å2 for the dimer);
- conservation of the interface residues, which EPPIC scores;
- experimental evidence such as size-exclusion chromatography, cross-linking or mutagenesis.
Small crystal contacts in 1BRS (several rows under 200 Å2 with P-values above 0.5) show what packing artefacts look like: small, few bonds, unremarkable hydrophobicity.
How should you use PDBePISA on a docked model?
A docked pose from ClusPro, HDOCK or HADDOCK, or an AlphaFold-Multimer prediction, has no crystal lattice. PISA simply analyses the coordinates you give it, so the CSS and assembly verdict are less informative than for a crystal. What remains useful:
- Interface area and residue counts to describe the size of the contact;
- Hydrogen bond and salt bridge lists to name the key residue pairs;
- ΔiG and P-value to say whether the contact is hydrophobic;
- Comparison across poses: run the top three to five poses and report whether the same residues recur.
Pair it with an affinity estimate (our guide to protein-protein binding affinity prediction covers PRODIGY) and, if you run molecular dynamics, check whether the PISA salt bridges survive in the trajectory with salt bridge analysis in GROMACS.
Which interface analysis tool should you use?
| Tool | Main question it answers | Input | Primary reference |
|---|---|---|---|
| PDBePISA | Interface area, solvation energy, bonds, likely stable assembly | PDB code or upload (PDB/mmCIF) | Krissinel and Henrick, J. Mol. Biol. 2007 |
| EPPIC | Is a crystal interface biological or a packing contact, using evolution | PDB code or upload | Duarte et al., BMC Bioinformatics 2012 |
| PRODIGY | Predicted binding affinity (ΔG, Kd) from interfacial contacts | Uploaded complex, with the two interactors chosen by chain | Xue et al., Bioinformatics 2016 |
| PDBsum | Pictorial summary of an existing PDB entry, including interface residue plots | PDB code | Laskowski and colleagues, EBI |
A defensible results section on a docked complex usually combines PISA for the description and one of PRODIGY or MD-based energies for strength.
Why does PDBePISA give wrong or empty results for my upload?
Only one structure appears and no interface is found
Cause: both partners share one chain ID, which happens with some docking outputs. PISA sees a single structure. Fix: rename the second partner’s chain (for example in PyMOL, alter chain A and resi 1-120, chain="B" after checking the residue range), then save and re-upload.
Dozens of interfaces appear for a two-chain model
Cause: the file carries a CRYST1 record copied from a template, so PISA builds symmetry mates and finds lattice contacts that do not exist in your model. Fix: delete the CRYST1 line from a docked or predicted model before upload.
Ligands, ions or waters show up as extra interfaces
Cause: PISA treats HETATM groups as separate structures, as in the MK1 rows of 1HSG. Fix: remove waters and any groups you do not want counted (in PyMOL, remove solvent and remove hetatm), unless a cofactor genuinely sits in the interface.
Results look odd after uploading an ensemble file
Cause: several MODEL blocks in one file, as in NMR-style or merged docking ensembles. Fix: split the ensemble and upload one model per file.
Numbers differ from another program’s buried area
Cause: PISA uses its own surface calculation with a 1.4 Å probe and enlarged atomic radii to approximate hydrogens, and it reports half the buried area. Fix: compare like with like, and state the program and version in your methods.
How do you report PDBePISA results in a thesis or paper?
A compact pattern that reviewers accept:
Methods: “Interfaces were analysed with PDBePISA (Krissinel and Henrick, 2007) on the top-ranked docked pose after removal of waters and heteroatoms.”
Results: give a small table per complex with interface area per side (or BSA, labelled as such), iNres for each partner, ΔiG, P-value, NHB and NSB, followed by the three to five most important residue pairs from Details with their distances. Then interpret, and say what PISA cannot tell you. PISA’s own citation page asks that results from the service be cited whenever they are published or presented.
For where interface analysis sits among the other skills you need, see the molecular docking learning path and the computational biology skills roadmap.
Frequently asked questions
Is PDBePISA free?
Yes. It is a free web service run by the Protein Data Bank in Europe at EMBL-EBI, with precalculated results for the whole PDB and interactive calculation for uploaded files.
What is a good ΔiG value in PDBePISA?
There is no universal threshold. Negative ΔiG with a P-value below 0.5 indicates a specifically hydrophobic interface. ΔiG ignores hydrogen bonds and salt bridges, so polar interfaces such as barnase with barstar can show ΔiG near zero and still bind extremely tightly.
Is the PISA interface area the same as buried surface area?
No. PISA’s interface area is half of the total buried surface area, because it divides the lost accessible surface by two. Double it to compare with papers that report BSA.
Can PDBePISA calculate binding affinity?
No. ΔiG is a solvation term, not a binding free energy, and PISA does not output a Kd. Use PRODIGY or an MD-based method such as MM/PBSA for an affinity estimate.
What does CSS 0 mean for a docked complex?
CSS reflects whether the interface takes part in an assembly PISA predicts as stable. For a docked model with no crystal context, a low CSS is common and does not by itself mean the pose is wrong. Lean on the residue-level details and an independent affinity or MD check.
Want the guided, hands-on version?
Our live Molecular Modeling & MD Simulations cohort bootcamp takes you from zero to running real docking and MD workflows, with a portfolio project for your grad-school applications.