A highly efficient blocked Gibbs sampler reconstruction of multidimensional NMR spectra
Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, PMLR 9:940-947, 2010.
Projection Reconstruction Nuclear Magnetic Resonance (PR-NMR) is a new technique to generate multi-dimensional NMR spectra, which have discrete features that are relatively sparsely distributed in space. A small number of projections from lower dimensional NMR spectra are used to reconstruct the multi-dimensional NMR spectra. We propose an efficient algorithm which employs a blocked Gibbs sampler to accurately reconstruct NMR spectra. This statistical method generates samples in Bayesian scheme. Our proposed algorithm is tested on a set of six projections derived from the three-dimensional 700 MHz HNCO spectrum of HasA, a 187-residue heme binding protein.