Enhancing Crystal-Structure Prediction with NMR Tensor Data

A method for selecting high-probability structures from numerous computer-generated crystal structures is described. This procedure evaluates structures by comparing computed NMR shifts for each predicted structure to experimental solid-state NMR data. Four carbohydrates are evaluated: methyl β- d -galactopyranoside, methyl α- d -glucopyranoside, methyl α- d -mannopyranoside, and methyl β- d -xylopyranoside. In these cases, 81.8% of the structures retained as probable fits by lattice energy comparisons are eliminated by the NMR criterion using tensor principal values. This analysis also ranks the correct structure as the best-fit for methyl α- d -glucopyranoside and usually places the correct structure among the top five in other cases. Isotropic shift comparisons are less successful in selecting structure. The NMR analysis is sufficiently sensitive to identify a 30° error in one torsion angle of the purported correct structure of methyl β- d -galactopyranoside. In this case, it is found that none of the 164 computer-generated structures match experimental data. The substances investigated experience only weak electrostatic fields; therefore, the NMR analysis chooses primarily by molecular conformation rather than lattice structure. NMR data thus provide a valuable independent selection criterion. The presence of strong electrostatic fields in polar samples can alter the results given here, and likely changes in the selection process are discussed.