Supplementary Figure 12: Structural accuracy and NMR DP scores are correlated for ²H,¹³C,¹⁵N-enriched, I(δ)LV methyl-protonated proteins.

Protein samples were generated using standard protocols²⁶. Backbone resonance assignments were determined using standard ²H-decoupled triple-resonance NMR experiments, as described elsewhere³. For each protein, 2000 decoy conformers, spanning a range of accuracies, were generated use the CS-Rosetta structure modeling program²⁷ together with these backbone chemical shift data. The resulting decoys were compared with reference structures determined by X-ray crystallography, which are deposited in the Protein Data Bank, by backbone RMSD (x-axis). DP scores (y-axis), comparing each decoy model with the NOESY peak list and chemical shift assignment list, were computed using the RPF-DP program^16,25. Backbone RMSDs were calculated by comparing the Rosetta decoy to the X-ray structure in well-ordered regions, which was defined consistently for all decoys using the the corresponding NMR structure solved using a fully-protonated sample, available from the PDB. (A) Protein GmR137, 78 residues, MW = 8.5 kDa (reference PDB ID 3CWI). (B) Protein HR3201A, 87 residues, MW = 9.9 kDa (reference PDB ID 3KW6). (C) Cold-shock protein A, 70 residues, MW = 7.4 kDa (reference PDB ID 1MJC). Most structures classified as “reliable”, with backbone RMSD < 4 Å from the reference structure, have DP scores > 0.73.