Determining the three-dimensional structures for large proteins using Nuclear Magnetic Resonance (NMR) poses a formidable undertaking. First, noise in multidimensional NMR data exhibits unknown spatial inhomogeneity due to systematic noise and local correlation. Second, there will be a large number of protons that resonate at similar frequencies. High dimensional NMR solves ambiguities due to similar resonance frequencies at the price of a lower sensitivity. Therefore, it is important to identify NMR spectra for a range of NMR signal dimensions.

The primary objective of the research presented in this talk is to develop a statistical technique for identification and characterization of multi-dimensional NMR spectra. Our statistical method takes a novel overall perspective: (1) It incorporates a preliminary step for separating the signal from the background using a method that adapts for sharp changes in the data and non-homogeneous signal; (2) The locations, widths and amplitudes of the NMR spectra identified above a noise level-dependent threshold are estimated using a computational efficient algorithm; (3) It detects mixtures of spectra to solve ambiguities due to protons with similar resonance frequencies; (4) It applies to all range of NMR spectra dimensions.