JOURNAL OF MAGNETIC RESONANCE 88,60 l-608 ( 1990) A Versatile Approach toward the Partially Automatic Recognition of Cross Peaks in 2D ‘H NMR Spectra GERARD J. KLEYWEGT, ROLF BOELENS, AND ROBERT KAPTEIN * Department of Chemistry, University of Utrecht, Padualaan 8, 3584 CH Utrecht, The Netherlands Received November 29, 1989; revised December 29,1989 NMR spectroscopy has established itself as a versatile technique for obtaining in- formation about the spatial arrangement of ever larger macromolecules of biological interest ( 1-3). The linchpin of any NMR-based structural study is the assignment of resonances to individual protons. Unfortunately, this amounts to a painstaking man- ual analysis and comparison of, typically, at least a dozen spectra (differing in nature of the experiment and experimental conditions, for instance, pertaining to solvent, pH, and temperature), which render it a rather time-consuming process. For this reason, workers in many laboratories the world over have begun to develop software tools that facilitate or automate this process, and, recently, these efforts have begun to bear fruit (4-8). As anyone who is working in this particular area will acknowledge, the initial data- reduction step (i.e., extracting a list of peak positions and, possibly, the corresponding intensities and/or peak volumes, from each of the spectra that are used) is of para- mount importance. A large number of computer programs which attempt to auto- mate this step have been published in the past few years ( 9-23). In our experience, such algorithms, which often employ considerations concerning the global and local symmetry of spectra, perform satisfactorily in spectra which have highly symmetrical cross peaks (such as those encountered in COSY spectra, where cross peaks may have approximate O2 or even DZd symmetry), provided that they do not display too much fine structure and that not too much overlap occurs. If these conditions are not met, however, one is almost bound to end up with a farrago of real peaks, noise peaks, and so-called “ghost peaks.” Moreover, categorical application of symmetry-based considerations where they do not fully apply may easily cause real peaks to be over- looked. Since CLAIRE ( 7)) our suite of programs for the automatic assignment of protein spectra, requires a HOHAHA and a 2D NOE data set (i.e., the positions and intensi- ties of all cross peaks), we were clearly in need of a versatile package of peak-picking software. We have therefore developed such a package and called it STELLA, short for synergistic approach toward the evaluation of local maxima in low-symmetry spectra. All programs were written in FORTRAN-77 and run on DEC (p)VAX com- * To whom correspondence should he addressed. 601 0022-2364190 $3.00 Copyright 0 1990 by Academic Press, Inc. All rights ofreproduction in any form reserved.