Biomedical Engineering Reference
In-Depth Information
Population 1: allele 1 more prevalent (p
1
= 0.8)
and disease more prevalent (K = 0.75). No
association between allele status and disease
status within this population:
Population 2: allele 2 more prevalent (p
1
= 0.2)
and disease less prevalent (K = 0.25) No
association between allele status and disease
status within this population:
2
= 0.
2
= 0.
χ
χ
Population 1 + Population 2.
Apparent association between allele status
and disease:
2
= 5.33; p-value = 0.021
(1 degree of freedom). There is a higher
proportion of disease (dark shading)
among the "1"s than among the "2"s.
χ
Figure 4.2
An example of spurious marker-disease association due to population stratification.
The two possible alleles (1 or 2) at the marker are denoted by the corresponding numeral. When
the allele is from an affected case, the numeral is shaded black; when the allele is from an
unaffected control, the numeral is unshaded (white).
K
is the population prevalence of the
disease (proportion of the population that is affected with the disease).
for the effects of stratification by computing an overall 'inflation factor' to adjust resulting
association statistics. The more recent EIGENSTRAT method [37] uses principal compo-
nents analysis to adjust for stratification and addresses the limitations of using a uniform
correction despite possible differences in allele frequencies across ancestral populations.
EIGENSTRAT is also amenable to using large numbers of markers such as would be avail-
able in a GWAS. Various studies have identified ancestry informative markers (AIMs) that
are expected to be especially useful to test for substructure [38 - 42]. Some commercial SNP
genotyping arrays, such as Illumina's DNA Test Panel (360 SNPs), also can provide SNPs
that are informative for assessing potential population stratification.
PROTOCOL 4.1 Generic Design and Quality Control for a Disease
Association Study Using Unrelated Case and Control Subjects
a
Equipment and tools
Association analysis software; for example PLINK [33].
b
•
