Foldiff · Structural comparison
Protein states
What changes between known structures of one protein? Compare residue positions, contacts and secondary structure.
One protein. Different states.
Start with the published example or add your structures. The tool checks sequences and numbering before calculating.
Measuring coordinate differences
Drag to rotate, scroll or pinch to zoom. Ligands are shown as sticks. This overlays states; it is not a transition trajectory.
Secondary structure map
Consistent coordinate calculationPyDSSP-C3: simplified DSSP rules, three classes. H includes α-, 3₁₀- and π-helices. C does not establish disorder. “?” does not mean structural loss. Assignments may differ from full DSSP and the 3D ribbon.
Where the chain moved
Cα displacement after alignment, Å. Select the chart or a row to highlight a residue.
Largest residue displacements
| Residue | Shift, Å | Structure A → B | Contacts + / − | Ligand ≤ 5 Å |
|---|
Table: consistent C3 assignment. Sequence mapping shows A / B numbers. Contact: Cα within the threshold, at least 3 positions apart. Ligand proximity does not establish binding.
Traceable examples
From domain motion to fold switching
Adenylate kinase shows domain motion; XCL1 shows two experimentally established folds; HcKCR1 shows dark and illuminated structures. Choose an example to see conditions, limitations and sources here.
Method and limitations
PDB and mmCIF, first coordinate model. Mapping uses residue numbering or explicit global alignment of observed sequences. Substitutions and ambiguous matches are rejected. At least 20 shared Cα and 80% of each chain are required; the explicit partial-coordinate mode allows 50%. Unmatched residues are listed. Conflicting known UniProt IDs are rejected.
Rigid alignment: Horn method, without reflection. RMSD describes geometry, not statistical significance. Cα contacts use shared residues of the selected chains. Changes within 0.5 Å of the cutoff are flagged. Anchor ranges use reference numbering.
Secondary structure: PyDSSP three-class rules, hydrogen-bond energy < −0.5 kcal/mol, masking proline donors and chain breaks. This is consistent coordinate assignment, not sequence prediction or full mkdssp. Beta bulges may be classified differently. N/Cα/C/O are required; incomplete backbones are “?”. By default, supplied protein chains in the first model are included; biological symmetry assemblies are not reconstructed.
H/C/E assignments describe this coordinate model. One NMR conformer does not describe an ensemble distribution; model counts are not state probabilities. Water and nonstandard residues are excluded. B-factors are not pLDDT. Verify constructs, conditions and provenance in publications.
State energies, equilibrium populations, transition rates, photocurrent and activity are not calculated. The mutation tab uses a single-chain backbone model without membrane, chromophore or partner interactions. Different numbering requires preparation before mutation inference.
PyDSSP · MIT · License · Horn, 1987 · Numerical checks and disagreements
Testing prediction sensitivity
Does the prediction hold across structures?
The same single substitution, the same model, multiple known states. Negative ΔΔG predicts stabilization; positive ΔΔG predicts destabilization.
From one prediction to a state comparison
Open the example, or choose structures in the first tab and enter your substitutions.
ThermoMPNN · kcal/mol
Prediction matrix
0.5 is an initial selection margin, not a measured model error or confidence interval. Changing it changes classification, not predictions.
Each point represents one structure; the line spans minimum to maximum predictions. It is not an uncertainty interval. Exact values are in the table.
Reading the result
Stabilizing direction retained: all uploaded states have ΔΔG ≤ −margin. Destabilizing: all ≥ +margin. Direction changes: predictions occur below −margin and above +margin. Near zero / indeterminate: other complete sets. Insufficient data: any missing state prediction. Agreement does not establish that a mutation is useful.
The model receives the backbone of one chain. Ligands, partner chains, pH and temperature are not modeled. An incomplete backbone or chain break blocks prediction for that state. Values describe model sensitivity to structure, not transition energy or state occupancy. Activity and α ↔ β switching probabilities are not predicted.
Example: real 4AKE and 1AKE coordinates, but ΔΔG values are model predictions, not measured data. Substitutions use a regular sampling rule: every 13th position starting at the fourth, to A/V/D, excluding identity substitutions. This checks calculation reproducibility, not state-specific accuracy, and is not a list of recommended mutations.