RMSIP for three projected PCA modes is lower between model halves than the mean for fit masks in 638 ubiquitin partitions.

We compared top-three Cartesian principal component subspaces across equal model halves and between two alignment masks for three deposited ubiquitin NMR ensembles. The analysis used chain A C-alpha coordinates, residues 1–70 for PCA, and linear projection of six rigid tangent directions. Across 638 partitions, every paired gap was positive: within-half mask agreement exceeded between-half agreement. Median between-half RMSIP was 0.388529, 0.678154, and 0.763543 for entries 1D3Z, 2K39, and 6V5D; the corresponding full-ensemble mask RMSIPs were 0.999052, 0.998688, and 0.999752. The result indicates greater sensitivity to model selection than to these two fit masks in these selected, related ensembles. It does not establish mode accuracy or biological independence. The partitions overlap, 2K39 and 6V5D share data, and no held-out experimental accuracy test or authenticated 6V5D refinement-group map was available.

ubiquitin · principal component analysis · RMSIP · NMR ensembles · model selection · alignment · mode subspaces