Automated assignment of simulated and experimental NOESY spectra of proteins by feedback filtering and self-correcting distance geometry

A new method for automatically assigning proton-proton NOESY spectra is described and demonstrated for simulated and experimental spectra of the proteins dendrotoxin K, α-amylase inhibitor tendamistat and the DNA-binding domain of the 434 repressor protein. The method assigns the NOESY spectrum and calculates three-dimensional protein structures simultaneously, using a list of proton chemical shifts and ³J(NHα) coupling constants. An ensemble of structures is iteratively calculated by self-correcting distance geometry from unambiguous and selected ambiguous NOESY cross peaks. New structure based filters recognize the correct constraints from the ambiguous cross peak list. For the first round of assignment neither a preliminary initial structure nor a sufficient set of unambiguous NOESY cross peaks is needed. The method can also be applied to cross peak lists containing hundreds of noise peaks. For an assumed tolerance of +/- 0.01 ppm in the chemical shifts of the peak positions, only about 10% of the NOESY cross peaks can be unambiguously assigned based on their chemical shifts alone. Our automated method assigned about 80% of all cross peaks with this chemical shift tolerance, and 95 to 99% of the assignments were correct. The average pairwise RMSD for the backbone atoms of the ten best final structures is about 1.5 Å in all three proteins and the previously determined NMR solution structures are always embedded in this structure bundle. We regard our method as a highly practical tool for automatic calculation of three-dimensional protein structures from NMR spectra with minimal human interference.