Protein backbone chemical shifts predicted from searching a database for torsion angle and sequence homology

Chemical shifts of nuclei in or attached to a protein backbone are exquisitely sensitive to their local environment. A computer program, SPARTA, is described that uses this correlation with local structure to predict protein backbone chemical shifts, given an input three-dimensional structure, by searching a newly generated database for triplets of adjacent residues that provide the best match in φ/ψ/χ1 torsion angles and sequence similarity to the query triplet of interest. The database contains 15N, 1HN, 1Hα, 13Cα, 13Cβ and 13C' chemical shifts for 200 proteins for which a high resolution X-ray (≤2.4 A) structure is available. The relative importance of the weighting factors for the φ/ψ/χ1 angles and sequence similarity was optimized empirically. The weighted, average secondary shifts of the central residues in the 20 best-matching triplets, after inclusion of nearest neighbor, ring current, and hydrogen bonding effects, are used to predict chemical shifts for the protein of known structure. Validation shows good agreement between the SPARTA-predicted and experimental shifts, with standard deviations of 2.52, 0.51, 0.27, 0.98, 1.07 and 1.08 ppm for 15N, 1HN, 1Hα, 13Cα, 13Cβ and 13C', respectively, including outliers