Developing and Evaluating a Novel SNP Haplotype Analysis Method Shaun Purcell1,2, Mark Daly1

1
Developing and Evaluating a Novel SNP Haplotype
Analysis MethodShaun Purcell1,2, Mark Daly1
Pak Sham21Whitehead Institute, Cambridge, MA.
2Institute of Psychiatry, London, UK.
3. Secondary test The power of the omnibus test
can be increased via an independent secondary
test. The secondary test assesses the correlation
between pairwise haplotype genetic similarity and
haplotype effect similarity. One definition of
genetic similarity is the number of matching
sites, either globally or local to a specific
SNP. Similarity of effect is defined as the
squared difference of the regression coefficients
estimated in the omnibus test. The secondary
test is implemented as a multivariate weighted
least squares regression of effect similarity on
genetic similarity. An empirical significance
value can be combined with the primary omnibus
significance value, by Fishers method of
log-transformation.
5. Research questions We have performed
preliminary coalescent simulations to begin to
address three broad questions When is
haplotype analysis advantageous compared to
single SNP analysis? Is an omnibus test
preferable to haplotype-specific tests? How well
does the secondary test perform?

• 1. Overview
• This poster
• describes a novel two-stage method of haplotype
  analysis, implemented in the software package
  whap
• reports a preliminary set of coalescent
  simulations
• shows that haplotype analysis, in the current
  simulated scenario, is more powerful than single
  SNP analysis

7. Results Omnibus versus haplotype-specific
tests There were no significant differences
between the power of the omnibus and
haplotype-specific primary tests, for both rare
and common CVs. With a greater number of
replicates and/or studying CVs rarer than 5, one
might expect differences to arise -- this will be
investigated further in more comprehensive
simulations. All following results use the
omnibus test. Single locus tests versus primary
and secondary haplotype analysis The secondary
test increases the power of haplotype analysis,
particularly for common CVs A direct
single SNP test of the CV is more powerful than
haplotype analysis however, after adjusting for
multiple testing (CV CORRECTED), the combined
primary secondary test is as powerful
Apart from at the CV, haplotype analysis is
generally more powerful than single SNP analysis
in this scenario, particularly after correcting
single SNP analysis for multiple testing

• 2. Background
• Traditional haplotype analysis compares estimated
  haplotype frequencies in cases versus controls. A
  new class of method Schaid et al (2002) Am J
  Hum Genet Seltman et al (2003) Genet Epi
  Sham et al (in press) Behav Genet treats the
  individual as the central unit of analysis. We
  propose a novel two-stage method
• The primary test is a regression-based analysis
  of association of haplotype and trait
  (quantitative or qualitative) with one regression
  coefficient per haplotype.
• The secondary test analyses the relationship
  between haplotype similarity and regression
  coefficient similarity.
• This method also allows
• between/within family partitioning for robust
  family-based tests
• optional modelling of genotype conditional on
  trait (selected samples, TDT)
• inclusion of covariates and moderator variables
  (epistasis, GxE interaction)
• permutation testing to generate empirical
  significance values

(ß 1 ß2 )2 ( 0.405 - 0.620 )2 0.462
6. Coalescent simulations We simulated a
candidate gene study design, with a single causal
variant SNP (CV) and 8 evenly-spaced marker SNPs
in a 20kb region. Simulations were performed
using the CoSi coalescent simulator developed at
the Whitehead Institute by Cat Foo Mark
Daly. In 100 replicate samples of 300
individuals, the CV explained 2 of the variance
in a quantitative trait. A recombination
hotspot was simulated in the center of the
region the CV was always at the 3rd position
(i.e. typically in the first of two haplotype
blocks). Methods are compared by
matched-pairs t-tests on the distributions of
log-transformed empirical p-values. Haplotype
analyses of all 9 SNPs (i.e. including the causal
variant) are labeled CV otherwise tests are
of 8 SNPs (i.e. excluding the causal variant) and
are labeled CV-. CVs with a minor allele
frequency less than 20 were labeled Rare,
which should index the CV as being relatively
recent. This gave approximately a 5050 split for
rare versus common CVs. Correction for multiple
testing in the case of single SNP analyses is
performed empirically, by comparing each observed
test statistic against the largest of the 8/9
test statistics generated in each of 5000
permutations.

• 4. An example
• An illustrative sample comprising 8 SNPs in a
  20kb region was simulated (see section 6). The
  results of the primary omnibus and secondary
  (global and local) tests are reported below
  (selected output from whap)

Hap Freq Alt Null ---
----- ------ ------ 22211121 0.396
0.000 1 0.000 1 22111121 0.173
0.090 2 0.000 1 21211221 0.110
-0.045 3 0.000 1 22122212 0.095
0.289 4 0.000 1 12211221 0.071
0.261 5 0.000 1 21222212 0.040
0.053 6 0.000 1 22111212 0.035
0.711 7 0.000 1 22211221 0.030
-0.294 8 0.000 1 22111221 0.026
0.373 9 0.000 1 12211212 0.024
-0.604 10 0.000 1 --- -----
------ ------ Primary omnibus analysis
p 0.004975 Secondary analyses global
p 0.0398 combined p 0.001886 (S 8.083
) loc 1 p 0.1542 combined p
0.006270 (S 2.174 ) loc 2 p 0.0895
combined p 0.003883 (S 3.687 ) loc 3 p
0.0049 combined p 0.000287 (S 21.725
) loc 4 p 0.0746 combined p 0.003303
(S 2.928 ) loc 5 p 0.0746 combined
p 0.003303 (S 2.928 ) loc 6 p 0.4179
combined p 0.014919 (S 0.297 ) loc 7
p 0.8706 combined p 0.027903 (S 0.010
) loc 8 p 0.8706 combined p 0.027903
(S 0.010 )
8. Summary The novel secondary test appears to
increase the power of haplotype analysis, such
that it can be as powerful as a single locus test
of the actual causal variant after correction for
multiple testing. Further simulations in
different scenarios are required to generalize
these results.