PARIS, a program for automatic recognition and integration of 2D NMR signals

Abstract

The quantitative evaluation of 2D maps is the requisite step for some of the most fascinating recent developments in NMR techniques such as, for example, the use of the full relaxation matrix over a series of 2D NMR maps corresponding to various mixing times leading to a more accurate description of 3D geometries of macromolecules ( I, 2)) and the analysis of the cross-peak fine structure based on either pattern recognition (3,4) or symmetry reduction (57) algorithms allowing a more extensive description of the coupling networks. However, due to the increasing complexity of the systems and particularly the growing number of proteins and nucleic acid fragments (Fig. 1) under study in many laboratories, manual extraction of the information present in the 2D NMR spectra, but dispersed in hundreds of peaks, is extremely tedious if not impossible. Computeraided approaches are therefore essential in any practical solution. Moreover, an automated 2D spectra analysis should be easily implemented on commercial computers, have a fast algorithm, even on the smaller computers normally used for controlling the spectrometer operations, and be reasonably efficient when operating under the difficult conditions usually encountered in the study of biological macromolecules (low signal-to-noise ratio, presence of overlapping peaks, and presence of artifacts). The automated recognition methods of Madi et al. (6) and of Glaser and Kalbitzer ( 7) are presently to our knowledge the only fully automated approaches for 2D spectra reported in the literature, besides Pfandler’s pattern recognition technique (3, 4) which is of a more global philosophy. Another very interesting possibility (8) relies on the prior knowledge of those individual 1 D signals which could give rise to cross peaks. This facilitates the task but is inherently restricted in scope, particularly for larger spin systems. The methodology we propose here is based on the identification of the local extrema, followed by a precise evaluation of the peak extent. Its main disparity with the preceding ones lies in its completely different approach for the peak extent definition module, which allows a much clearer separation of the cross peaks in crowded regions. Moreover, since in many applications such as NOESY and HOHAHA spectra the fine structure present in the cross peaks merely complicates the analysis, we devised an algorithm that is able, at will, to handle cross peaks with or without fine structure. The program may roughly be broken into two consecutive operations, namely the extraction of the local maxima (peak picking) and the determination of the in-