Systems Biology (Methods in Molecular Biology)

3.3.3 NMR Data
Postprocessing

NMR data postprocessing is a necessary step of metabolomics

pipeline to extract useful information related to the state of

biological system. This step helps to avoid sources of variation in

the data, such as dilution effect, subtle changes in chemical shifts,

line-widths, and baseline across series of spectra, which can interfere

with the outcome of the statistical analysis, leading to false

deductions.

NMR data postprocessing usually includes exclusion of

non-informative regions, binning, normalization, scaling, and

data export for subsequent multivariate statistical analysis.

1. Remove theregions in thespectra thatcontain only noise and/or
   exogenous peaks. Therefore, exclude the spectral regions outside
   the window 0.5 (including TSP signal) and 9.0 ppm and those
   containing the residual water (δ4.7–5.0 ppm) and drug peaks.


2. Reduce the dimensionality of data splitting the p-JRES spectra
   into small segments (bins or buckets) with variable widths
   ranging from 0.01 to 0.04 ppm to ensure that each bin con-
   tains the same signals throughout all the spectra. If local peak
   shifts across series of spectra are still observed, compress groups
   of bins into single bins or alignment of the spectra. Then,
   integrate the signal within each bin (seeNote 5).


3. Normalize the binned spectra by applying the Probabilistic
   Quotient Normalization (PQN) [10, 11] method to make
   spectra comparable:
   (a) Set the total spectral area of every spectrum to 100.
   (b) Calculate as a reference spectrum the median spectrum
   (median of each variable/bin area) of healthy group
   samples.
   (c) Calculate the quotient between the area of each spectral
   bin of the considered spectrum and that of the
   corresponding bin in the reference spectrum.
   (d) Calculate the median of all the quotients.
   (e) Divide all the variables of the considered spectrum by the
   median quotient.
   (f) Repeatsteps c–efor all spectra.


4. Scaling the data by applying the generalized log (g-log) trans-
   formation [12, 13] to make the variables within spectra
   comparable:
   (a) Estimate the g-log transformation parameter (λ) by the
   maximum likelihood method using a set of five replicate
   measurements.
   (b) Obtain these five replicates from a single homogeneous
   pool of fecal water samples from healthy and pathological
   patients. Process the replicate spectra as described above

Metabolomics and Clinical Needs 331