Abstract

A well known problem with several types of multi-array NMR techniques is the visual representation of the results in terms of physico-chemical parameters associated with individual spectral peaks. This regards, for example, the Diffusion Ordered Spectroscopy (DOSY) in which the non-spectral parameter is the diffusion coefficient and ROSY (Relaxation Ordered Spectroscopy) for which it is the relaxation time. In all such cases, the transformation of the original data set to a suitable final 2D graph (i.e., the DOSY and ROSY transforms, respectively) is conceptually difficult to manage and current approaches are either little more than sketches (evaluating only spectral points corresponding to peak tops) or prone to characteristic noise-induced artifacts (fitting of spectral cross sections with little or no signal).

We have developed a Bayesian approach to this problem which is very efficient computationally and physically meaningful, and which gives very satisfactory, artifact-free results. Applied specifically to the DOSY and ROSY data sets, it leads to what we call the BDT (Bayesian DOSY transform) and BRT (Bayesian Rosy transform) algorithms.

The BDT algorithm has been implemented and is now being tested on real experimental data. Ongoing work aims at improved, Bayesian handling of overlapping spectral peaks belonging to different sample components.