Previous Topic Back Forward Next Topic
Print Page Frank Dieterle
Research TutorialsResearch Tutorials MetabonomicsMetabonomics MethodsMethods Pre-processingPre-processing Probabilistic NormalizationProbabilistic Normalization
About Me
Ph. D. Thesis
Research Tutorials
    Biology and "-omics"
    Definition of Metabonomics
      In-life phase
      Analytical Techniques
        Spectral Exclusion Regions
        Peak Shifts
        Probabilistic Normalization
      Data Analysis
      Biochemical Interpretation
    Applications in Drug Development
Downloads and Links
Site Map
Print this Page Print this Page

Probabilistic Normalization of Spectra

The integral normalization is not valid, if the integral of a spectrum does not represent the overall concentration of the corresponding sample. This can happen, if extreme amounts of single metabolites are excret or if drug metabolites are excreted. In this case single peaks dominate the integral of the corresponding spectrum. The integral normalization is "fooled" and incorrectly down-scales the corresponding spectrum proportional to the area of the dominating peaks. This problem is shown in the first figure.

The second figure shows the same spectra after an optimal normalization. It is visible that both spectra differ only by few peaks besides of the dominating glucose peak. The optimal normalization is based on a probabilistic normalization. This approach scales the spectrum in way, that most metabolites have the same concentration. This procedure is based on the fact that metabonomic changes only influence few metabolites, whereas overall concentration changes influence all metabolites the same way. Thus the optimal normaliztion is the normalization, which calculates a most probable concentration of the sample based on the majority of metabolites. More details can be found in F. Dieterle, A. Ross, G. Schlotterbeck, H. Senn: Method for processing a set of spectra, particularly NMR spectra, Patent Application.

Two NMR spectra after integral normalization. The sample with the red spectrum contains extreme amount of glucose. The huge glucose peak, which is marked by a magenta arrow, dominates the area below the spectrum. Therefore the integral normalization does not scale the spectra to the same overall concentration.


The same spectra after an optimal normalization. Besides of the glucose peak only few differences of the spectra are visible, which are marked by magenta arrows. The normalization shown is based on probabilistic normalization proposed by Dieterle et al..


Page 163 © Frank Dieterle, 03.03.2019 Navigation