pyRMSD

What is pyRMSD?

pyRMSD calculates an all-versus-all root mean square deviation matrix for a matching ensemble of protein conformations. It is designed for structures that contain the same atoms in the same order, such as frames from a molecular dynamics trajectory or alternative models of one protein.

RMSD measures coordinate displacement after optimal superposition:

$$\operatorname{RMSD} = \sqrt{\frac{1}{N}\sum_{i=1}^{N}\lVert \mathbf{x}_i-\mathbf{y}_i\rVert^2}$$

where $N$ is the number of selected atoms. The value is reported in Å.

How to use pyRMSD online

Run pyRMSD online by uploading or fetching at least two matching PDB conformations, selecting alpha carbons or all atoms, and choosing a QCP, Kabsch, QTRFIT, or no-superposition calculator. ProteinIQ returns every unique structure pair as a table, plus the condensed NumPy matrix and pyRMSD summary statistics.

Input requirements

Matching atom counts are necessary but not sufficient. Two unrelated structures can contain the same number of atoms while pairing the wrong coordinates. For proteins with different sequences, residue numbering, missing regions, or chain layouts, use USAlign or the RMSD calculator, which provide structure-matching workflows.

Atom selection

Calculator

The fitted calculators solve the same geometric objective through different numerical algorithms, so results should be very close apart from floating-point variation. OpenMP changes execution parallelism, not the scientific definition.

Results

For $n$ conformations, the table contains $n(n-1)/2$ unique pairs.

pyRMSD stores only the upper triangle because an RMSD matrix is symmetric and its diagonal is zero. pyrmsd_condensed_matrix.npy preserves that compact one-dimensional form for NumPy workflows. pyrmsd_statistics.json contains the source matrix minimum, maximum, mean, variance, skewness, and kurtosis where available.

Interpreting an RMSD matrix

There is no universal threshold for a similar protein conformation. RMSD depends on protein size, atom selection, flexible termini, loops, domain motion, and whether fitting was enabled. A 2 Å all-atom RMSD is not equivalent to a 2 Å alpha-carbon RMSD.

The matrix is most informative as a relative map within one consistent ensemble:

• Low-RMSD blocks: Groups of conformations that may represent the same structural state.
• High values in one row: A possible outlier, transition structure, or malformed frame.
• Two or more blocks: Candidate conformational substates for clustering or representative selection.
• Broad all-atom values with tighter CA values: Similar backbones with differing side-chain conformations.

Global RMSD is sensitive to localized motion. A flexible terminus can increase the value even when the protein core is stable, and domain movement can dominate over local structural agreement. Atom selections should therefore be chosen before comparing experiments, and every matrix being compared should use the same selection and fitting method.

For a single reference-versus-model comparison with explicit residue matching, the RMSD calculator is easier to interpret. pyRMSD is the better choice for complete pairwise analysis of a topology-consistent ensemble, including conformations exported from molecular dynamics.