RMSD calculator

How to calculate protein structure RMSD online

Root-mean-square deviation (RMSD) measures the distance between corresponding atoms in two structures, reported here in ångströms. For equally weighted atoms, it is the square root of the mean squared distance after the selected fit. A lower value means closer agreement for that atom selection; compare coverage and use the same fitting settings before comparing scores between jobs.

Upload or fetch a reference structure and one or more comparison structures, choose the atoms and residue regions to fit and measure, then select Calculate. ProteinIQ matches corresponding coordinates, performs an optimal rigid-body fit, and returns global RMSD, displacement statistics, matched coverage, a residue profile, multi-model summaries, structure overlays, and the transformation applied to each comparison.

For a verified example, fetch 1UBQ as the reference and 1D3Z as the comparison. Keep the default settings:

RMSD atom selection: Alpha carbons (CA)
Alignment atom selection: Same as RMSD selection
Compare by: Structure
RMSD weighting: Uniform
Multi-model structures: First model only
Fit residue numbers: All residues
Measurement residue numbers: All residues
Automatic alignment: On

The current RCSB coordinate files produce:

Reference: 1UBQ
Comparison: 1D3Z:model_1
RMSD: 0.521 Å
Measured atoms: 76
Fit atoms: 76
Reference coverage: 100%
Comparison coverage: 100%

1D3Z contains 10 NMR models, so first-model mode also reports that only model 1 was used. Choose All model pairs to compare every 1D3Z model.

The 0.521 Å result is reproducible for the current RCSB PDBx/mmCIF files and the settings shown above. It compares all 76 matched Cα atoms in 1UBQ with model 1 of 1D3Z; it is not a general threshold for deciding whether two unrelated proteins have the same fold.

Input

PDB IDs can be fetched directly from RCSB. Fetched entries use PDBx/mmCIF, which preserves long chain identifiers and structures that cannot be represented in legacy PDB format.

How atoms are matched

In Structure mode, atoms correspond only when all of these identifiers match:

chain ID + author residue number + insertion code + record identity + atom name

For protein ATOM records, the residue name is not part of the key, so atoms shared by a point mutation can still be compared. For HETATM records, record identity includes the residue name. A ligand atom in LIG therefore cannot collide with a protein atom or an atom from a differently named heterogen at the same author position.

In Chain mode, the calculator first separates the structures into chains and then compares every reference chain with every comparison chain. Chain ID is omitted from the atom key inside each selected pair, which allows chain A to be compared with an otherwise corresponding chain B.

In Mapped chains mode, only explicit pairs such as A:X,B:Y are compared. This avoids the Cartesian set of chain comparisons when the biological correspondence is already known. Use (blank) for an empty chain ID. A mapping to a chain that is absent from a selected model produces a warning for that model pair.

Alternate coordinate locations are not counted twice. The calculator keeps blank shared atoms, chooses one occupancy-preferred alternate conformation per residue, and reports removed duplicate records as a warning. Atom pairs are never discarded as outliers after fitting, so every reported matched atom contributes to the RMSD.

This is an identifier-based RMSD calculator, not a sequence or structural alignment search. If chain IDs, numbering, or sequences differ substantially, use USAlign, which finds a structural correspondence before reporting RMSD and TM-score.

Settings

Choosing an atom selection

All-heavy and all-coordinate modes require matching ligand, solvent, and modified-residue identifiers. A low matched coverage means the reported RMSD describes only the shared subset.

Atomic-mass weighting reads element symbols from the coordinate records. If an element is missing or unsupported, the comparison stops with an error rather than assigning an arbitrary mass. Uniform weighting avoids that dependency and is the conventional choice for Cα and backbone RMSD.

Fit on one selection and measure another

The alignment selection can differ from the RMSD selection. For example, align on alpha carbons and measure all heavy atoms to ask how well the full atomic model agrees after the protein fold is fitted. In that case, the reported RMSD is minimized for the Cα fit selection, not necessarily for the heavier measurement selection.

The two residue fields are independent. For example, fit residues 10:80 to establish a stable domain frame and measure residues 81:120 to quantify movement in a second domain. Both filters use author residue numbers from the coordinate files. Both structures must use the same numbers for each filter; use USAlign when their numbering differs.

Results

Global results

Automatic alignment requires at least three non-collinear matched fit atoms. The separately fitted transformation can then evaluate one or more matched measurement atoms. A comparison that cannot meet those minima is reported as a warning instead of producing a misleading or geometrically ambiguous value.

Reading RMSD and coverage together

RMSD has meaning only for the atom pairs that were actually measured. Coverage shows how representative that shared set is.

Comparisons are most defensible when they report the atom selection, fit selection, weighting, matched atom count, and both coverage values alongside RMSD. Values from different selections or different matched subsets are not directly interchangeable.

Per-residue RMSD

The per-residue table applies the same global transformation and then groups squared atom displacements by reference residue. This highlights flexible termini, loops, binding-site changes, and local side-chain differences that may be hidden by one global value.

Per-residue RMSD is not a separate fit for each residue. Every residue is evaluated in the coordinate frame established by the global or chain-level alignment.

The Displacement tab plots these per-residue RMSD values in matched-residue order. Tooltips preserve the author residue label, both chain IDs, both residue names, and the matched atom count. Profiles above 5,000 display points are reduced with ordered minimum and maximum points from each interval; the complete values remain in the per-residue table. The numeric results also include atom-level mean, median, P95, and maximum displacement, so the plot and distribution answer related but distinct questions.

Multi-model summary

The Model summary tab groups all successful model pairs for each structure or chain pair. It reports the minimum, mean, median, maximum, population standard deviation, and the model pair with the lowest RMSD. In first-model mode each group contains one comparison; in all-model mode it provides a compact ensemble summary.

Files

The Run log is a concise reproducibility record. It does not include uploaded coordinates, raw exceptions, environment values, storage details, URLs, or infrastructure diagnostics.

The structure viewer opens overlays in displacement coloring. The derived reference file stores 0 in the B-factor field. Fitted comparison files store the matching per-residue RMSD in that field; atoms outside the measured or matched set also store 0. These are visualization annotations in derived files and do not preserve the source B-factor values.

Fitted mmCIF input remains mmCIF. A PDB input is returned as mmCIF when a chain ID cannot fit the one-character legacy PDB field or when a displacement annotation cannot fit the fixed-width PDB B-factor field.

Each run returns at most 100 overlay coordinate files and 100 MB of overlay coordinate content. If that download limit is reached, the complete numerical, per-residue, summary, and transformation results are still returned. Requests are also limited to 1,000 structure, model, or chain comparisons and 250,000 per-residue rows so unusually large combinations fail with an actionable message instead of exhausting the browser.

What is protein structure RMSD?

Protein structure RMSD is a coordinate-distance metric for a defined set of corresponding atoms. After superposition, it summarizes the squared distances between those atom pairs as one value in ångströms. Lower values mean closer agreement for that exact correspondence, selection, and fit, but RMSD does not discover which residues should correspond.

For $n$ corresponding atoms with weights $w_i$, ProteinIQ calculates:

$\mathbf{R}$ is the optimal proper rotation calculated from the matched fit atoms, so mirror images cannot be fitted by reflection. $\mathbf{t}$ translates the rotated comparison into the reference coordinate frame. With uniform weighting, every $w_i$ is 1. With atomic-mass weighting, $w_i$ is the mass of the matched element.

The fit minimizes the weighted squared displacement of the selected fit atoms. A separate measurement selection is evaluated after that same transformation, which is why fitting on Cα atoms and measuring heavy atoms can produce a heavy-atom RMSD that is not itself minimal.

There is no universal RMSD threshold for structural similarity. Global RMSD is sensitive to protein size and to a few large deviations, including flexible termini, loops, and domain motion. Matched coverage, per-residue RMSD, and the biological question determine whether the global value is informative.

Which structure comparison tool should I use?

FAQ

Should I use Cα, backbone, or all-atom RMSD?

Use Cα RMSD for overall fold comparison, backbone RMSD for main-chain geometry, and heavy or all-coordinate RMSD only when the structures have closely matching atomic composition. Compare values only when atom selection, fitting, weighting, and coverage are consistent.

Why is matched coverage low?

The structures differ in chain IDs, author residue numbering, insertion codes, missing atoms, or atom names. Chain mode can reveal corresponding chains with different IDs. For broader correspondence problems, use USAlign.

Why did a comparison not produce a row?

The warning panel states the reason. Common causes are an invalid coordinate file, no matching selected atoms, fewer than three matching atoms, or collinear fit atoms that cannot define a unique three-dimensional rotation.

Can this calculate ligand RMSD?

Yes, when the ligand is present in both structures with matching chain, residue, insertion-code, and atom-name identifiers. Choose All heavy atoms and use chain mode when the ligand has its own chain. Symmetry-corrected ligand RMSD is not applied, so chemically equivalent atom permutations require a docking-specific evaluation workflow.

What happens to alternate locations and multiple models?

For alternate locations, one occupancy-preferred residue conformer is selected and duplicate coordinates are removed. Multi-model files use model 1 by default; All model pairs evaluates every reference model against every comparison model.

Should automatic alignment ever be disabled?

Disable it only when coordinates are already in the same reference frame and the absolute displacement is the quantity of interest. Without alignment, rigid translation or rotation contributes directly to RMSD.