Biology Reference
In-Depth Information
burden, and presumable any associated markers,
are at low concentration) and speci
samples are labeled as either Melanoma (M) or
Normal (N) and clearly separated in this plot.
The clear separation of the data set into two
distinct groups leads to the possibility of using
this set of mass spectrum peaks as a potential
classi
c (will distin-
guish malignant from nonmalignant cases)?
A recent publication describing highly sensi-
tive detection of melanoma based on serum pro-
teomic pro
ling
11
provides a very instructive
example of using proteomics to attack this
problem and the associated multivariate anal-
ysis. In this approach the human proteome in
patient serum was studied with surface-
enhanced laser desorption/ionization/time-of-
s mela-
noma status. Techniques beyond the scope of
this chapter can be used to calculate the
optimum separator between two prede
er (i.e., diagnostic) for a patient
'
ned
classes; this
figure corresponds to the diagonal
blue line in
Figure 6
B just to the right of the Mela-
noma group. The authors then test their classi
flight mass spectrometry (SELDI-TOF-MS). The
experimental design included 60 samples from
patients with a con
er
using an independent data set. It is always
important to test any MVA model using inde-
pendent data to validate the model.
rmed diagnosis of MM (30
stage I/II, 30 stage III/IV) and a control group
of 48 healthy volunteers, age and sex matched.
Proteomic analysis was performed as described
in the paper. Univariate statistical analysis was
used to identify the peaks in the MS data that
showed a clear difference between the control
and melanoma samples (
Figure 5
).
*
In other
words, the protein peaks that were differentially
expressed between the melanoma and control
were identi
STUDY 2: DETECTION OF HEART
DI
SEASE BY METABOLOMIC
S
Coronary heart disease (CHD) is the leading
cause of death in the United States for men and
women and has many known risk factors (age,
gender, diabetes, smoking, blood pressure,
genetics, etc.).
12
There is a gold standard for
diagnosis: coronary angiography is a method
that involves threading a long, thin tube (called
a catheter) through an artery or vein in the leg
or arm and into the heart designed to evaluate
the heart arteries. Unfortunately this technique
is both highly invasive and expensive, which
signi
ed. The top 24 of the resulting 112
variables were then selected for further analysis.
The data matrix,
X
, is then constructed as dis-
cussed in the
first part of this chapter [
Eq. (2)
],
with the 108 rows representing the patients and
the 24 columns representing the area of the cor-
responding mass spec peak. Hierarchical clus-
tering and a heat map provide a useful
visualization of the data structures being
analyzed.
Figure 6
A shows this visualization in
which (normalized) values are represented on
a color scale for visualization purposes. There
are clear signs of structure (clusters)
cantly limits its use.
With the current
imperative for reducing
health care costs,
finding alternatives that
address these limitations is of high interest.
A recent paper has explored the potential for
NMR metabolite pro
in the
ling as an alternative
noninvasive method for the identi
data, which motivates further analysis.
In order to look for a potentially actionable
difference between the healthy and diseased
tissue, a PCA analysis was done on the reduced
MS data. A score plot based on the
cation of
CHD.
13
The 600-MHz
1
H-NMR spectra of
human sera from patients with severe CHD
(triple vessel disease [TVD] patients;
n
first two prin-
cipal components is shown in
Figure 6
B. The
36)
and patients with angiographically patients;
n
¼
30) can be compared visually in
Figure 7
A.
Patients in the TVD group had signi
¼
cant coro-
*
A nonparametric Mann-Whitney U test was used (
P
<
nary artery disease (de
ned as a reduction of
2).
11
0.05/fold change
>
Search WWH ::
Custom Search