Biomedical Engineering Reference
In-Depth Information
Before lowess normalization
After lowess normalization
6
5
4
2
0
0
-2
-5
-4
0
5
10
15
6
8
10
12
14
16
A
A
8.1
The two panels represent MA plots of data from a two-color array
before and after lowess normalization. On the
x
axis (
A
) is the average
of the log ratio and on the
y
axis (
M
) is the log ratio of the two samples.
It is apparent that before normalization there is an intensity-dependent
bias in the measurements. The data cloud is crescent-shaped and
there is bias in the low-intensity data towards up regulation. After
normalization, the data cloud is centered around no change (0 on the
y axis) and symmetrical around it.
plot used for two- color arrays (two samples on the array) is the MA plot
(Smyth and Speed, 2003). In this plot, 'M' on the
y
axis is the log or the ratio
and 'A' on the
x
axis is the average intensity (Fig. 8.1). Another useful plot
for single-sample arrays is the box plots of each array's intensity before and
after normalization. You can clearly distinguish arrays with different means
and standard deviations and whether or not normalization was effective.
8.3.6 Outlier fi ltering
The last step before analysis is to identify outliers. If after normalization
you observe one or two arrays that do not behave or look remotely close in
intensity to the other arrays (scaling factor too high), or notice some area in
an array that looks like an artifact, then the wise choice is to eliminate them
from the analysis. Due to the usually small number of replicates, outliers can
really wreak havoc in the analysis. One graphical representation very useful
to identify outliers is the plots obtained by the principal component analysis
(PCA). PCA takes data and plots it against the three main orthogonal axes
of intrinsic variation found in the data. Some analysis software has a three-
dimensional (3-D) plot that you can manipulate to view all the angles. In
this type of plot, it is easy to identify microarrays that are outliers, or if there
