Practical guide · updated October 17, 2026
How to Compare WT and Mutant Enzyme Structures
A useful WT–mutant comparison makes the chosen atoms, chemical assumptions and reference explicit. Start with a defined experimental question, then document the coordinate differences without turning them into an activity claim.
Try the free comparison exampleNo account, upload or GPU run needed for the example.
1. Check the inputs before measuring
Keep the original WT and mutant PDB or mmCIF files. Record whether each is an experimental structure or a model prediction, which construct was used, and which chains form the assembly being compared. A label such as K186M is insufficient if the two files use different constructs or residue numbering.
Review the substrate identity, stereochemistry, formal charge and modeled state. Record any cofactor and its chemical state, and metal species and counts for the complete modeled assembly. If these differ between WT and mutant inputs, explain why before attributing a coordinate difference to the mutation. Leave unresolved chemistry as an explicit question.
Map the mutation to the correct structure chain and residue. Sequence positions and deposited residue numbers may differ. EnzymeAI does not yet provide automatic sequence-to-structure alignment; verify that correspondence before uploading a measurement recipe.
2. Define the atoms and the reference
Write down the exact atom identity for every measurement: chain, residue number, insertion code, component and atom name. Keep the model and alternate location choices explicit. Avoid combining mutually exclusive alternate conformations into a single geometric interpretation.
Choose WT as the reference initially. In the workbench, select Compare structures, upload the complexes and choose the reference pose. The 3D explorer supports atom clicks, residue search such as A:186, an atom selector within the residue, and explicit Use as atom A/B/C controls. Search follows deposited numbering and does not align the proteins. Check atom identities in each file; identical-looking labels do not establish equivalent chemistry. Ligand atom names can differ between export pipelines, so verify the correspondence rather than assuming a match.
3. Measure distances and angles
A distance uses two selected atom coordinates and is reported in ångströms. An angle uses three atoms A–B–C, with B as the vertex, and is reported in degrees. Choose atoms that test your particular reaction hypothesis. There is no universal distance or angle threshold that identifies a better enzyme.
EnzymeAI reports the raw measurement and its difference from the selected reference. In the public stored WT/K186M example, O16–PG changes from 4.90 to 3.77 Å, giving mutant minus WT of −1.13 Å. O16–PG–O3B changes from 152.00 to 167.18°, giving +15.18°. The decimal places describe the exported-coordinate calculation, not its experimental accuracy.
Switching to K186M as reference reverses the two shared differences. If you enter a minimum or maximum, record the reaction-specific evidence behind it. A criterion describes your interpretation of those coordinates; the software does not validate the mechanism or infer a rate constant from it.
4. Keep missing atoms unavailable
A substitution can remove an atom used in the WT measurement. In the K186M example, the Lys186 NZ atom is absent from Met186, so its NZ–PG contact cannot be measured in the mutant. EnzymeAI reports it as unavailable. It does not insert zero or silently substitute a different atom.
Other missing measurements may reflect incomplete exports, numbering mismatches, alternate-location choices or absent ligands. Inspect the explanation before drawing a conclusion. If you define a chemically different contact in the mutant, label it as a separate measurement instead of presenting it as the original atom pair.
5. Review confidence separately
Keep predicted-coordinate geometry separate from prediction-confidence evidence. For supported Boltz outputs, the workbench accepts a native confidence summary after you confirm that it belongs to the same run and pose. Filename checks help identify mismatches; they do not cryptographically bind the report to a structure.
A whole-complex confidence summary does not establish the accuracy of a chosen active-site contact. Review the local evidence available from the original prediction and record anything absent. The public demo supplies no confidence summary. For file naming and output fields, see the Boltz 2.2.1 output documentation.
6. Save enough to reproduce the result
Export the measurements CSV, measurement recipe and analysis report from the full workbench. Retain the source structures, construct and chemistry notes, chosen reference, prediction settings and any confidence files. The report records source filenames and exact selected coordinates; the built-in atlas example also includes verified file hashes.
The standalone HTML report opens offline and can be printed to PDF. Report coordinate changes as observations to examine experimentally. Two selected prediction samples do not establish equilibrium populations, catalytic rates or improved substrate specificity.
Start with a reproducible example
The free demo shows full WT and K186M protein coordinates from two stored Boltz predictions. It verifies their hashes and checks every original pocket atom against the corresponding full structure. Explore the substrate, ATP and modeled ions in 3D alongside each measurement. Try the reference switch and download the report, then use your own complexes in the design workbench.
Structure comparison runs locally in your browser. Sequence analysis and external prediction have different data flows; review where your inputs go before using project data.