Biology Reference
In-Depth Information
and data storage or warehousing is an important
point for its durability.
simultaneous detection of fragments and adducts
by theMS. Deconvolution algorithmswere devel-
oped to handle these signals properly, but such
a process may be insuf
Resolution Tuning, Noise Filtering, and
Mass Features Detection
Tuning data resolution is a preliminary and
cient due to the presence
of overlapping peaks when analyzing complex
samples. When performing large-scale metabolo-
mics of complex biological matrices, it is likely
that some chromatograms will not be integrated
optimally. An overall manual inspection of the
automatic procedure is therefore recommended
to ensure a suitable processing of the data.
in
uential step of standard data processing
work
ow. Data binning/bucketing is a common
preprocessing procedure that corresponds to the
partitioning of the data. It involves the integra-
tion of signal intensities according to intervals
of constant or adaptive width size to generate
features. A close relationship with the data
dimensionality must be emphasized by reducing
the data resolution.
As a subsequent step, a
Spectral Features Alignment
The proper correspondence of the variables
across multiple samples constitutes another crit-
ical prerequisite to ensure the validity of further
analysis. The alignment of mass features
detected in different samples aims at the removal
of shifts between samples for a given compound.
Multiple factors
d
which include temperature
changes, mobile phase pH, pump pressure
filtering procedure is
usually applied to raw data to reduce random
analytical noise and correct systematic drifts.
Signal smoothing algorithms such as Gaussian,
moving window, or Savitzky-Golay
filters are
filtering.22
22
Noise
threshold can then be estimated to distinguish
true peaks from false positive ones. Baseline
shifts in the raw data are detected and the shape
of the drift is assessed, such as with a polynomial
function, before appropriate subtraction. Noise
often applied for noise
uc-
tuations, sample carryover, and column clogging
and degradation
d
may lead to differences
affecting both the
m/z
and the retention time
values. Shifts in the
m/z
dimension are usually
adjusted using appropriate MS calibration, but
drifts of retention times may be dif
filtering and baseline correction are usually per-
formed to ensure proper peak detection by elim-
inating analytical artifacts.
Peak detection can then be performed
through the selection of signals corresponding
to local maxima or
cult to
handle.
25
Many alignment techniques have
been developed to minimize run-to-run shifts
while preserving chemical selectivity.
26
The binning of data in the chromatographic
dimension constitutes the simplest way to
proceed. Misalignments are then limited to the
bin boundaries, as peaks can be alternatively allo-
cated to neighboring bins.
27
But such a process
induces a decrease of initial resolution. Addition-
ally, variations in the retention times are often
nonlinear
28
and more sophisticated methods are
needed.
Warping alignment procedures are performed
by shrinking or stretching the temporal axis to
achieve an optimal overlap across samples.
Segments of adjustable width are de
fitting more elaborated math-
ematical models, such as Gaussian distributions.
The apex and the in
ection points are used for
area integration.
23
Constraints on the peak
shapes in the chromatographic dimension
24
and criteria of minimal intensity, area, or
signal-to-noise ratio are usually applied to
distinguish true peaks from noise. Several
parameters generally must be adjusted to match
the chromatographic characteristics of the data.
For each compound in the sample, a single ion
is expected to be detected when applying
soft ionization. However, the detection of
biomarkers may be more complex, due to the
ned and
sequentially aligned.
29
The size of the segments
plays therefore a central role for the adequacy of
Search WWH ::
Custom Search