[1] Both the ShiftX program and the ShiftX web server make use of pre-calculated, empirically derived chemical shift tables relating 1H, 13C and 15N chemical shifts to backbone torsion angles, side chain orientations, local secondary structure and nearest neighbor effects.
These tables were derived using data mining techniques from a large database of reference-corrected protein chemical shifts called RefDB.
ShiftX differs from other protein chemical shift calculation techniques in that it blends both empirical observations with classical or semi-quantum mechanical approaches.
Recently, substantial improvements to the performance of ShiftX were achieved by using machine learning methods to better integrate protein structure features (including solvent accessible surface area) and local or nearest-neighbor interactions.
ShiftX2 achieves correlation coefficients between experimentally observed and predicted backbone chemical shifts of 0.98 (15N), 0.99 (13CA), 0.999 (13CB), 0.97 (13CO), 0.97 (1HN), 0.98 (1HA) with corresponding RMS errors of 1.12, 0.44, 0.51, 0.53, 0.17, and 0.12 ppm.