**3. Materials and Methods**

The NMR and crystallographic structures analyzed in this study consisted of all (41) NMR structures and all but one (40) crystallographic structure listed in the "community resource" described by Everett et al. [35], which also outlines standardized methods used by the NESG for solving crystallographic and NMR structures. All but one of the NMR structures (1XPV) analyzed here were refined using CNS [37,38] and/or XPLOR-NIH [39]. MD simulations were performed on a randomly selected set of 12 targets from the community resource, using the conditions indicated in Table S1, Supplementary Materials. Most simulations used the OPLS [46] forcefield, but several simulations were performed with the AMBER99SB [47,48] forcefield as well.

MD simulations were initiated using crystallographic structures retrieved from the Protein Data Bank (PDB, [49]) with the identifications (IDs) listed in Table S1, Supplementary Materials. Simulations were prepared with Schrodinger's Maestro GUI made available as part of the Desmond [50] software package (which also ran MD simulations), using Na<sup>+</sup> or Cl<sup>−</sup> ions to achieve electrical neutrality and the TIP4PEW water model. In order to avoid artifacts due to truncation of the simulated constructs and facilitate parameterization in AMBER99SB, the terminal amino acid residues present in the coordinate sets obtained from the PDB were capped. Simulations ran for up to 36 ns (following default relaxation/minimization protocols), with snapshots recorded every 14.4 ps (up to 2500 snapshots). Re-parameterization of each simulation to use the AMBER99SB force field was performed using Desmond's Viparr utility. Most simulations were run at room temperature (generally defined for each protein by the temperature at which NMR experiments used to solve the protein's structure were performed. For all proteins in this study the temperature was very nearly 300 K in order to mimic the conditions in both the NMR tube and during (room temperature) crystallization. Some simulations were also performed at 100 K to mimic conditions obtained during cryo-cooled x-ray diffraction experiments. Simulations were ran both with and without substituting methionine (MET) for the seleno-methionine (SeMET) residues found in crystallographic structures. Dangling ends of protein chains absent from the crystallographic coordinates deposited in the PDB were not filled in computationally but rather were omitted from each simulation.

Initial parsing and visualization of each trajectory were performed using VMD [51]. A simple trajectory rescuer was used prior to initial parsing in VMD for simulations that turned into hung processes. Reformatting was completed for the multi-structural PDB file output from VMD into a multi-model format suitable for further analysis. THE-SEUS [30] superimposed MD trajectories prior to a coordinate variance calculation and the MATLAB [52] implementation of the FindCore Toolbox superimposed NMR ensembles. Calculation of coordinate uncertainties (calculated as coordinate variances) from FindCore superimposed NMR ensembles used the FindCore Toolbox and calculation of coordinate uncertainties and variances from THESEUS superimposed NMR ensembles and MD trajectories was also performed in MATLAB.

Friedman's test [36] is a non-parametric analog of ANOVA with repeated measures used here to compare whether coordinate uncertainties, variances, and B-factors are significantly different for different atom types. Application of Friedman's test proceeded as follows. For each residue in each structure, backbone heavy atom coordinate uncertainties, coordinate variances, or B-factors (depending on the analysis performed) were ranked (from 1–4). For each structure, the resulting ranks were tabulated with columns (treatments) corresponding to a heavy atom type (amide N, Cα, carbonyl carbon, and carbonyl oxygen) and one row (block) for each residue, and the resulting table was subjected to Friedman's test, which compared column averages (average rank by heavy atom type, averaged on a per-structure basis). MATLAB scripts tabulated B-factor and coordinate variance/uncertainty data for analysis via Friedman's test and subsequent multiple comparisons, which were also performed in MATLAB. MATLAB was also used to calculate an F-score measuring the relative uncertainties, variances, or B-factors of carbonyl oxygen atoms in a given residue (Equation (1)):

$$F = \left(\mathfrak{u}(\mathcal{O}) - \mathfrak{u}(N)\right)^2 / \left(\mathfrak{u}(\mathcal{O}) - \mathfrak{u}(N)\right)^2,\tag{1}$$

where *u*(.) denotes the coordinate uncertainty, variance, or B-factor of the given atom and *O*, *N*, and *C*′ are the carbonyl oxygen, amide nitrogen, and carbonyl carbon atoms, respectively.'

**Supplementary Materials:** The following are available online: Table S1: Parameters/Input for MD Simulations. Table S2: Average Ranks of Backbone Atom B-Factors for Crystallographic Structures. Table S3: Average Ranks of Backbone Atom Coordinate Uncertainties for Theseus Superimposed NMR "Ensembles". Table S4: Average Ranks of Backbone Atom Coordinate Uncertainties for Find-Core Superimposed NMR "Ensembles". Table S5: MD Simulation Results. Table S6: Average Ranks of N, C', Cα, and Cβ B-Factors for Crystallographic Structures. Table S7: Average Ranks of N, C', Cα, and Cβ Coordinate Uncertainties for Theseus Superimposed NMR "Ensembles". Table S8: Average Ranks of N, C', Cα, and Cβ Coordinate Uncertainties for FindCore Superimposed NMR "Ensembles". Table S9: Average Ranks of N, C', Cα, and Cβ Coordinate Variances in Theseus Superimposed MD Trajectories. Table S10: Average Ranks of N, C', Cα, and H Coordinate Uncertainties for Theseus Superimposed NMR "Ensembles". Table S11: Average Ranks of N, C', Cα, and H Coordinate Uncertainties for FindCore Superimposed NMR "Ensembles". Table S12: Average Ranks of N, C', Cα, and H Coordinate Variances in Theseus Superimposed MD Trajectories. Figure S1: Distribution of average ranks of coordinate uncertainties, variances, and B-factors of N, Cα, carbonyl C, and Cβ atoms. Figure S2: Results of Friedman's Test and subsequent multiple comparisons analysis with Cβ atoms. Figure S3: Distribution of average ranks of coordinate uncertainties, variances, and B-factors of N, Cα, carbonyl C, and Cβ atoms. Figure S4: Results of Friedman's Test and subsequent multiple comparisons analysis with amide H atoms.

**Author Contributions:** Conceptualization: D.A.S.; Data curation: D.A.S., C.R., A.R. and J.R. Formal analysis: D.A.S., C.R., A.R. and J.R. Investigation: D.A.S., C.R., A.R. and J.R. Methodology: D.A.S. Software: D.A.S., C.R., A.R. and J.R. Writing—original draft: D.A.S., C.R., A.R. and J.R. Writing review and editing: D.A.S., C.R., A.R. and J.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** While working on this project during the Summer of 2019, Christopher Reinknecht received a stipend from the Garden State—Louis Stokes Alliance for Minority Participation (GS-LSAMP), an NSF funded program.

**Data Availability Statement:** Input files used to run the MD simulations analyzed in this paper, the resulting (superimposed) MD trajectories, and all the scripts used to perform the analyses reported here are all archived on Zenodo, doi:10.5281/zenodo.4323630.

**Acknowledgments:** The authors graciously acknowledge Adrian Roitberg for his insights into the results presented by DAS at the ACS Spring 2014 Meeting in Dallas. The authors also thank Gaetano Montelione for his insights into NMR-based structural determination as well as for his helpful comments and constructive discussion, and John Chodera for his suggestion to publish a previous iteration of this study on bioRxiv. An award of Assigned Release Time for research from the Office of the Provost of William Paterson University of NJ facilitated completion of this work.

**Conflicts of Interest:** The authors have no conflicts of interest to report.

**Sample Availability:** Samples of the compounds are not available from the authors.
