Biology Reference
In-Depth Information
is advisable to plot CV after missing value imputation and data
transformation in order to check outliers within the replicates and
remove them from analysis.
σ
=
CV
k
×
100
x
In the case of 2-DE gels, all of the outliers/non consistent
spots should be manually supervised in the image analysis software
before removing it, since erroneous or mismatched spots are not
unusual.
The large proteomic datasets usually shows a linear dependence
between mean and standard deviation, as a consequence of the
digital acquisition process. Furthermore, in almost all cases, the
low number of replicates led to a nonnormal distribution of the
values of each variable.
A simple log 10 or cubic root transformations performs
adequately for controlling the mean/SD dependence and also for
normalizing the variables in most of the cases.
3.5 Data
Transformation
3.6 Univariate
Statistics
The use of this group of statistical tests, which compares the vari-
ables one by one, represent the classical approach for distinguish-
ing the possible changes in the protein abundance as a consequence
of the application of the different treatments. These tests are suit-
able to define differential proteins, biomarkers, etc. but lack the
capacity of distinguishing between groups, multi-variable trends or
hidden patterns as multivariate statistics do (see below).
The normality and homogeneity of variance (homocedasticity) of
the samples should be tested prior to applying univariate statistics.
Shapiro-Wilk test for nonnormality can be applied on R statistical
software using the shapiro.test{stats} function. Bartlett test can be
applied for testing if samples are from populations with equal vari-
ance. In R this test is available in the bartlett.test{stats} function.
3.6.1 Normality
and Homocedasticity
1. Student's
t
-test
This procedure is classically used to test the null hypothesis that
the means of two normally distributed populations are equal
(i.e. if Actin abundance is the same in the control and the treated
sample). We can compare only two treatments using this test
and no block effects can be estimated. Furthermore, and strictly
speaking, this test can only be used if the variances of the two
populations are assumed to be equal. If this assumption is
dropped the form of this test is called Welch's
t
-test. In R we
have the function t.test{stats} that can be used for computing
both Student's and Welch's
t
-test.
3.6.2
Parametric Tests
