Population Stratification Analysis in Genome-Wide Association Studies - Mathematical Approaches to Polymer Sequence Analysis and Related Problems - page 192

Tabl e 9. 9 Inflation factor ()
values calculated in
EIGENSTRAT chi-squared
statistics after correction
using 1 PCs (K1) to 10 PCs
(K10) in the UCI sample and
in the CATIE-NIMH sample.
In the CATIE-NIMH sample
we show the results both
without or with removal of
outliers

CATIE-NIMH CATIE-NIMH

UCI (COutliers) (Outliers)

K1 1.063 1.067 1.098

K2 1.01 1.067 1.036

K3 1.008 1.046 1.036

K4 1.005 1.041 1.035

K5 1.002 1.04 1.036

K6 1.001 1.038 1.034

K7 1.001 1.038 1.034

K8 1.003 1.037 1.033

K9 1 1.037 1.033

K10 1 1.037 1.033

Tabl e 9. 10 Comparison of most significantly associated SNPs
in chi-squared statistics of EIGENSTRAT after correction with
3 or 9 PCs, in the UCI sample
ANSTP ASFP NANSFN

K3 11 17 5

K9 7 21 15

K3 and K9 5 11 6

cause a loss or gain of significantly associated SNPs and then to genes associated
with disease (Table 9.10 ). In particular, higher the lambda value higher is the cor-
rection factor for chi-squared statistics. Given a fixed threshold for the significance
(p-value<10E 4), the number of significantly associated SNPs decreases with an
increased value of inflation factor.

9.4 Conclusions

Case-control studies are hampered by PS that can occur in populations and can lead
to significant associations being detected at loci that have nothing to do with dis-
ease. There is currently no consensus about the effectiveness of a specific strategy
that allows to correct for PS. Our goal, studying different populations and com-
plex diseases, is to find an efficient strategy that allows us to correct our ethnically
mixed samples and thus to avoid false-positive genotype-phenotype associations.
Our current study shows an application of different methods to measure stratifica-
tion when genotyping thousands of genetic markers in two American case-control
samples composed of different ethnic groups. The study, involving PCA and MDS
identification of different subpopulations in our samples, provided additional insight
into the substructure of American populations and differences among various ethnic
groups that may impact our understanding of the genetics of complex diseases. We
also emphasize the importance of controlling substructure in the ascertainment of