March 11- 16, 2001. Orlando, Florida, USA, 42nd ENC

The ABC of Metabonomics-Automated Baseline Correction

Antony Williams, Sergey Golotvin, Eugene Vodopianov, ACD/Labs
John Shockcor, Center for Metabonomics Research, Imperial College, University of London

Abstract

The optimal automated baseline correction algorithm remains elusive but the hunt continues to become ever more necessary. Nowadays hundreds of NMR spectra are generated on a daily basis in automation using NMR equipment provided in walk-up environments. It is common for the data analysis to require accurate integration for quantitation purposes and this requires high quality spectra be generated prior to automated analysis. One of the primary sources of error in any quantitative measurements in NMR spectra is a distorted baseline which can arise due to numerous electronic artifacts including the "dead time" problem of pulsed NMR, non-linearity of the filter phase response, the discrete nature of the Fourier transform, instrumental instabilities and other miscellaneous reasons. Often these problems are more extreme for spectra acquired in an LC-NMR run or at low concentrations since solvent suppression commonly has to be applied. Although some baseline problems can be avoided by adjusting acquisition parameters during the run-time of the NMR experiment and exploiting digital filtering and oversampling, post processing data treatment offers a more general way of correcting baseline distortions. Most popular approaches include reconstruction of the first points of the fid and approximation of the baseline in the frequency domain using Fourier series, polynomials and functions of special form. On the other hand, the majority of NMR desktop software allows the user to manually set the points belonging to the baseline and interpolate between them using analytical functions to completely model the baseline. While the results can often be good enough the method requires manual intervention and cannot be used for batch processing. On the other hand the quality of automated procedures is rarely sufficient when the baseline has severe distortions. The failures are generally due to both inadequate types of analytical functions used for modeling and poor recognition of the baseline.

In our efforts to optimize capabilities for the diverse needs of NMR spectroscopists who "want-it-all" in a desktop NMR processing package, it has become obvious that one of the more tedious tasks in processing NMR data is baseline correction. Preferably this correction can also be performed without manual intervention to support the high-throughput environments of today's laboratories. We will report here on a recently developed baseline correction method which continues to pursue the elusive optimum. We will illustrate the successes using examples from LC-NMR and analysis of samples to support metabonomics research.