Each model removal changes the 1D3Z correlation matrix more than tail inclusion at a 0.03 Å floor.

Across 302 deposited ubiquitin NMR models from PDB entries 1D3Z, 2K39, and 6V5D, exhaustive single-model deletion changed the 70-residue correlation matrix more than the core-versus-full alignment mask in 1D3Z at a 0.03 Å variance floor: all 10 removals exceeded the full-data tail-mask change, by at least 4.74705 times. The corresponding counts were 13/116 for 2K39 and 5/176 for 6V5D. The seven 1D3Z removals meeting the 0.2 sign-change size cutoff affected 92 unique residue pairs; no 2K39 or 6V5D removal met it. Refitting alignment changed the matrices further, but the counts above the tail effect were the same with fixed full-data alignment. These are observations on selected deposited coordinates, not independent biological observations, confidence intervals, or evidence of physical interactions. Shared experimental data and refinement dependence limit comparisons between entries, and the analysis does not establish which correlation matrix is physically accurate.

ubiquitin · NMR ensembles · protein correlation · model deletion · structural alignment · variance weighting · sensitivity analysis