NMR Ensemble Analysis

An NMR structure "ensemble" is not a single model — it is a bundle of conformers, typically 10–40 structures, each consistent with the experimental restraints (NOE distances, dihedral angles, RDCs). The spread of the bundle is the primary indicator of precision: tight bundles indicate well-determined, rigid regions; loose bundles indicate disorder or sparse restraint coverage.

This page explains the scientific foundations of the ensemble analysis tools in synth_pdb.ensemble.

Why Ensembles?

X-ray crystallography produces a single, time-averaged structure. NMR is unique in that it inherently produces a probability distribution of structures. The ensemble:

Represents the genuine conformational heterogeneity of the molecule in solution
Encodes uncertainty — regions with few restraints have higher spread
Provides a direct view of internal dynamics on the µs–ms timescale

The two key quantities characterising an NMR ensemble are:

Quantity	Unit	What it measures
Pairwise RMSD	Å	Coordinate precision — how similar are the models geometrically?
DAOP (S(φ)+S(ψ))	dimensionless (0–2)	Dihedral precision — how consistent are the backbone angles?

Dihedral Angle Order Parameter (DAOP)

Introduced by Hyberts et al. (1992), the DAOP measures the circular variance of backbone dihedral angles across all ensemble members.

For a dihedral angle \(\alpha\) measured in \(N\) models:

\[S(\alpha) = \left| \frac{1}{N} \sum_{k=1}^{N} e^{i\alpha_k} \right|\]

This is the length of the mean resultant vector on the unit circle, ranging from 0 (completely disordered) to 1 (perfectly ordered).

\[S(\phi) + S(\psi) \begin{cases} \geq 1.8 & \text{well-defined residue} \\ < 1.8 & \text{disordered / poorly restrained} \end{cases}\]

The threshold of 1.8 is the PDBStat convention used by the BMRB deposition system (Tejero et al. 2013).

Coordinate RMSD

The pairwise RMSD matrix (size \(N \times N\) for an \(N\)-model ensemble) encodes all structural differences. Key derived quantities:

Mean pairwise RMSD: overall precision of the ensemble
RMSF per residue: standard deviation of each residue's position after superposition — directly comparable to B-factors and to MD RMSF
Medoid: the model with the lowest mean RMSD to all other models — the most "representative" structure

All RMSD calculations in synth_pdb.ensemble use Kabsch superposition (via synth_pdb.geometry.superposition) to remove rigid-body translation and rotation before computing coordinate differences.

Quality Thresholds

The following thresholds are taken from the PDBStat paper (Tejero et al. 2013) and are used by QualityAssessment:

Metric	Excellent	Acceptable	Poor
Backbone RMSD	< 0.5 Å	0.5–1.5 Å	> 1.5 Å
Heavy-atom RMSD	< 1.0 Å	1.0–2.5 Å	> 2.5 Å
Fraction well-defined	> 85%	60–85%	< 60%

IDP Ensembles

For Intrinsically Disordered Proteins (IDPs), these thresholds do not apply — a "good" IDP ensemble should have high RMSD and low DAOP in the disordered regions by design. The spread reflects genuine structural disorder, not poor data.

Validation of IDP ensembles requires comparing ensemble-averaged observables (PRE rates, SAXS profiles, chemical shifts) to experiment rather than applying rigid RMSD thresholds.

See Science: Intrinsically Disordered Proteins and the IDP Ensemble tutorial.

References

Hyberts, S. G. et al. (1992). The solution structure of eglin c based on measurements of many NOEs and coupling constants. Protein Science, 1(6), 736–751.
Tejero, R. et al. (2013). PDBStat: a universal restraint converter and restraint quality analyzer for protein NMR structures. J. Biomol. NMR, 56(4), 337–351.
Clore, G. M. & Iwahara, J. (2009). Theory, practice and applications of paramagnetic relaxation enhancement for the characterization of transient low-population states of biological macromolecules. Chem. Rev., 109(9), 4108–4139.