Agriculture Reference
In-Depth Information
phenotypic states allowing inferences on the effect of allele variation on
phenotype. There are several approaches to association studies depend-
ing on some previous analysis of the genotypic data and themainobjective
of the study. Those preliminary analyses include Hardy
Weinberg equi-
librium (HWE), missing genotype data, haplotype analysis, population
strati
-
cation (structure) determination, and genotypic data analysis. Devi-
ations from HWE can be due to inbreeding, population strati
cation, or
selection. One of the most important of these preliminary analyses is the
population structure determination.
The simplest way to do an association is with a biallelic marker and a
categorical trait. The statistical signi
cance of the resulting contingency
table can be estimated with a Pearson
X
2
test. For each quantitative trait,
the observed phenotypic data has a certain distribution. If the data has a
normal distribution, a student
'
s
t
-test can be used to determine the
signi
cance of the difference in means. For multiallelic markers and
heterozygous individuals, there is an increase in dimensionality,
because of an increment in genotypes and the possibility of genetic
dominance effects. For that situation, analyses of variance are preferred
(Fenoll et al. 2002). However, the interpretation of the ANOVA
'
s null
hypothesis of equality of means does not have a proper interpretation.
What should be done is to break down the hypothesis into easily
describable pieces.
The resulting contrast should correspond to relevant genetic effects that
can be probed on an individual marker basis or based on sequence
variants. The genetic effect can be tested with haplotype analyses in
which combinations of genotypes of different markers are contrasted.
Haplotype analysis is a powerful tool to detect variants with low fre-
quency. Nevertheless,
t-
tests and linear statistical models like ANOVA
rely on the normal distribution of data to estimate signi
cance. Deviation
from normality is common in association studies. A phenotypic popula-
tion distribution with a long tail could be due to experimental or genetic
effects, and it may be necessary to apply deconvolution to deal with
pattern of variation. The focus of the AM analysis is in the central part of
the distribution. Likewise, the focus of AMis usually on the alleles that are
in Hardy
Weinberg equilibrium.
Because of this, a threshold for allele frequency in the population is
often used to remove markers with rare alleles or alleles with very low
frequencies in the population. An imbalance in allele frequency might
lead to spurious results based on unequal variances between genotypic
classes for the phenotypic trait and subdivision of the population into
uneven samples. The use of nonparametric methods like Kolmogorov
-
-
-
Smirnov or Mann
Whitney tests will give robustness to those
