Biomedical Engineering Reference
In-Depth Information
(a)
(b)
Strong dye-
flip effect
Time effect
Pairing of the samples
Outlier
8.2
Principal component analysis (PCA). Panels (a) and (b) are a
rendering of PCA of the arrays against three main axes accounting for
the variation in the data. Frequently, the cloud of data points is evenly
distributed but occasionally, one can uncover patterns in the data that
require further attention for analysis. On Panel (a), we observe that for
this two-color array experiment there is a strong dye-fl ip effect (when
the experimenter switched the dyes of the two samples applied on the
array); we also have an outlier array which does not cluster with any
other. In Panel (b), we observe a time effect as the sample order along
the
x
axis corresponds to the order in which they were collected and
processed. There is also a pairing of the samples as the two treatments
are subjected to the same time effect. In this case, a paired t-test type
of analysis would be better to analyze the data as the variance between
the treatments is smaller than the variance due to the time effect.
is some variation in the data that needs to be taken into account such as a
dye bias, time effect, and operator effect ( Fig. 8.2 ).
8.4
Analysis
There are many different methods that can be employed to analyze nor-
malized data, depending on the experimental design and the questions that
are posed. This section will review the major types of analysis that can be
performed.
8.4.1 Class comparison
One of the most popular analysis methods, known as class comparison,
involves fi nding groups of genes, SNPs, samples, etc., that distinguish between
two or more classes. An example is identifi cation of groups of expressed
genes that distinguish between tumor and normal cells. Another example
is mutations that distinguish between drug responders, non-responders and
adverse responders.
