Accuracy and robustness of three-way decomposition applied to NMR data

Three-way decomposition is a very versatile analysis tool with applications in a variety of protein NMR fields. It has been used to extract structural data from 3D NOESYs, to determine relaxation rates in large proteins, to identify ligand binding in screening for lead compounds, and to complement non-uniformly recorded (sparse) spectra. All applications so far concerned experimental data sets; it thus remains to address questions of accuracy and robustness of the method using simulated data where the correct answer is known. Systematic tests are presented for relaxation and NOESY data sets. Mixtures of real and synthetic data are used to allow control of various parameters and comparisons with correct reference data, while working with input that is as realistic as possible. The influence of the following parameters is evaluated: signal-to-noise, overlap of signals and the use of a regularization procedure within the algorithm. The main criteria used for the evaluation are accuracy and precision. It is shown that deterioration of accuracy is indicated by internal checks such as decrease of precision. Both with relaxation data and when interpreting NOESY spectra, three-way decomposition exhibits a robust behavior in situations with severe signal overlap and/or poor signal-to-noise, e.g., by avoiding false positives in the NOE shapes of NOESY decompositions. As a complement to this study, three-way decomposition is compared to other methods that achieve the same type of results.