High-Throughput Analysis of Microdissected Tissue Samples - High-Throughput Image Reconstruction and Analysis - page 254

11.6.3 Sources of Variation When Studying Microdissected Material
The sources of variation in the qRT-PCR process include variation in the dissection
process and technical variation in the downstream molecular analysis. Variation
due to microdissection is an important factor to consider since only a few thou-
sand cells out of the entire samples are analyzed. Our experience, however, is that
dissection variance is relatively small if the process is carefully performed. In gen-
eral, for established protocols such as qPCR or qRT-PCR, the variation of the
downstream analysis itself (technical variation) is negligible. However, as a quality
control measure, three qPCR replicates are usually performed on each sample and
the average of the replicates is used for analysis.
11.6.4 Comparisons of Gene Expression Between Two Groups
Comparisons of the DC T or DDC T between groups or conditions can usually be
done with a two-sample t-test, whether the researcher is dealing with microdis-
sected material or not. This test assumes that the outcome variables are normally
distributed with the same variance in the two groups. Although the t-test is
not sensitive to departures from these assumptions, it is wise for investigators
to check whether these assumptions hold. Alternative statistical tests should be
considered if there are marked departures from normality or there is evidence for
different variances in the two groups. For example, the nonparametric Wilcoxion
rank sum test does not assume that the data in the two groups follow a normal
distribution. Further, the Welch t-test does not assume equal variance in the two
groups.
In many analyses, the expression levels for many genes will be compared across
groups. Thus, it is not obvious what P-value is required to be confident that a
particular gene expression comparison is statistically significant. In most studies,
whether they are based on LCM samples or nonmicrodissected material, a P-value
smaller than 0.05 is used as a benchmark for statistical significance. However,
when there are a large number of genes being compared, such a criterion will
produce a large number of false-positive results. For example, if an investigator
is comparing gene expression values of 100 genes across two groups, they should
expect five genes to be classified as ''significantly'' different across the two groups
just by chance alone. This is an inherent problem with performing multiple statis-
tical tests in the same analysis. Various approaches have been proposed to control
for these false positive results. One strategy, called a Bonferroni correction, is to
adjust the significance level by dividing by the total number of comparisons made.
For example, if one is comparing 50 genes across groups, a comparison is con-
sidered statistically significant if the P-value is less than 0.05/50=0.001. Such a
comparison guarantees that the chance of generating a false positive result is be-
low 5%. Although such an approach adequately controls the overall false-positive
rate, in many situations, it may have very little power (i.e., have a low probabil-
ity of detecting a meaningful difference). This may be a particular problem when
the number of genes being examined is large (say above 10 genes). An alterna-
tive approach is to control the type of false discovery rate (FDR) rather than the
probability of any false positive result. Specifically, this approach controls the pro-
portion of false positive results at a particular probability. For example, controlling