Wga Of Obesity
WGA of Obesity
Analytical Strategies and Examples
22/9/2006
1
Overview
The initial focus will be on quality control and a
summary analysis of individual SNPs, to be followed
by more detailed and sophisticated ones involving
hotspots and multiple SNPs and genes.
Research papers will be written for thorough
treatise on specific problems and findings. Note
items under “further analysis” in the following are
there only because of their labour-intensive nature
but may well be topics reported in separate papers.
I can make a note of discussion points and circulate
this later
22/9/2006
2
Descriptive analysis of traits
Descriptive analysis includes summary
statistics and graphics.
A decision has to be made regarding
outliers.
The non-genetic determinants of these
traits will be summarised or referred.
22/9/2006
3
Single locus analysis
Quality control including call rates
Hardy-Weinberg equilibrium (HWE)
Coding according to minor allele and
genetic models
Case-control comparison
22/9/2006
4
Multilocus analysis on regions
of interest
Linkage disequilibrium (LD) and
population characteristics
Haplotype analysis and covariate
adjustment, gene-environment
interaction
22/9/2006
5
Further analysis
Genotyping error
Sensitivity analysis
Population substructure
Multistage design and analysis, joint analysis
Gene-based analysis, Hotelling’s T2 statistic
and database extraction
Analysis of pathways
Meta-analysis with earlier reports and data
from other sources.
Retrospective versus prospective methods
22/9/2006
6
Implementation
The workhorse would be SAS/GENETICS or Stata, possibly in
conjunction with customised programs in C/C++, Fortran or S-
PLUS/R. This was only owing to the myriad of non-standard,
individually written, statistical genetic software available. The SAS
system is quite reliable, stable and available on both Windows and
Linux, along with Stata and S-PLUS/R, and some programs in C/C++,
Fortran. A by-product is customised software to be distributed.
The procedures in SAS/GENETICS were mainly designed for genetic
association studies of both population and family data, ranging from
single-locus analysis including allele/genotype frequency calculation,
Hardy-Weinberg equilibrium tests, genomic control to multilocus
analysis including LD measures, haplotype estimation, as well as
case-control association tests and adjustment for multiple testing. It
has the added advantage of being able to utilise a variety of standard
statistical procedures available. The organisation of these procedures
is such that all intermediate statistics and test results can be stored
as databases.
22/9/2006
7
More information
Detailed reviews of the advantage and
disadvantages of genetic analysis by individual
computer programs and on general statistical
packages have been given (Zhao and Tan 2006,
Hum Genomics, 2006, Curr Bioinformatics).
The following section describes two examples
showing prototype programs for the EPIC study of
400 controls on ~250k SNPs as well as analysis of
diabetes using chromosome 20 data from Ashkenazi
and four UK populations.
One might suggest comparison of results from the
SAS and Stata platforms used, but customised
programming will be required for Stata.
22/9/2006
8
EPIC 400 analysis
The extensible, modular SAS programs
shown in Figure 1 run without changes on
both Windows and Linux systems, and are
shared by all users who have access to the
database.
The results of HWE tests could be re-used
for several traits including BMI, HBA1c with
or without adjustment for covariates such as
age, sex.
The accommodation of both discrete and
continuous traits is also straightforward.
22/9/2006
9
Flowchart of the EPIC 400
Analysis
descrip.sas
phenotype.sas
qqplot.sas
regress.sas
multtest.sas
epic.sas
calls.sas
genotype.sas
hwe.sas
haploview.sas
HaploView
map.sas
snp.sas
EMSEMBL
22/9/2006
10
Chr 20 data
Apart from genotype coding which was programmed in SAS,
the analysis was done exclusively in Stata by Jian’an. Besides
Stata’s own facility, HWE tests and multiple testing can be
done via publicly available routines. It has additional difficulty
with covariates than framework currently available in the
literature.
Exclusion criteria include: i. SNPs with no coordinates (5) or
overlap of coordinate but unable to locate from Enembl or
Entrez Gene (1); ii. allele frequencies < 1% in combined data
(270); iii. HWE in cases <0.00001 or controls <0.0001 in
combined data (190); iv. Call rate < 80% in individual cohort
or <90% in combined data (86); v. p<0.0001 between stages
1 and 2 allele frequencies in either cases or controls in
individual cohort or combined (37). A total of 589 SNPs were
removed from the dataset (out of 4546=4544 at phase 1 +
751 at phase 2).
22/9/2006
11
Hotelling’s T2 statistic
It can be derived as a score statistic,
discussed in the context of staged design,
and can be used to test for multiple SNPs
while accounting for correlations them.
In the case of linkage equilibrium the score
statistic involving several neighbouring loci
reduces to the sum of individual chi-squares.
In fact the sum statistics have been
exclusively studied by Ott and associates.
22/9/2006
12
Joint analysis of data from two
and three stages
An earlier work was due to Lowe et al. (2004) Gene Immun
which included a multistage framework called stopping for
futility. The paper nevertheless focused on the design.
A following paper by Skol et al. (2006) Nat Genet showed joint
analysis is more powerful than replication study. It uses test
statistics involving several disease models and does not take
into account correlations between SNPs.
A recent approach was described by Lin (2006) Am J Hum
Hered, which also accommodates dominant/recessive/additive
recoding on genotypes with significant levels obtained from
Monte Carlo simulation. Other work by Lin and colleagues was
haplotype analysis of a variety of study designs including case-
cohort.
22/9/2006
13
SAS/GENETICS
The module consists of procedures ALLELE,
CASECONTROL, FAMILY, HAPLOTYPE, HTSNP,
INBREED, PSMOOTH, TPLOT which implement
allele, genotype and haplotype frequency estimation
and tests for differences appropriate for unrelated
individuals and family data. In conjunction with
other procedures such as LOGISTIC, GLM,
GENMOD, MIXED, PHREG it provides comprehensive
and integrated environment for analysis. The
database, graphics, programming, Internet facility,
among others, is well-documented.
22/9/2006
14
Summary Statistics
. count
7686
. table cohort case sample, col row
| SAMPLE and CASE
| -------- a --------
-------- b --------
COHORT | 0 1 Total 0 1 Total
----------+-------------------------------------------
0 | 1,343 1,343
1,343
1,343
1 | 2,124 376 2,500 2,137 363 2,500
|
Total | 2,124 1,719 3,843
2,137 1,706 3,843
22/9/2006
15
The age distribution
. histogram age, normal by(case sample)
0, a
0, b
5
.
0
0
i
t
y
s
n
1, a
1, b
e
5
D
.
0
0
40
50
60
70
80
40
50
60
70
80
age
Density
normal age
Graphs by CASE and SAMPLE
22/9/2006
16
The distribution of BMI
. histogram bmi, normal by(case sample)
0, a
0, b
.
3
.
2
.
1
0
i
t
y
s
n
1, a
1, b
e
D
.
3
.
2
.
1
0
20
30
40
50
60
20
30
40
50
60
participant's body-mass index in kg/m^2 at 1HC
Density
normal bmi
Graphs by CASE and SAMPLE
22/9/2006
17
Further summary
. sum bmi
Variable | Obs
Mean Std. Dev. Min Max
-------------+--------------------------------------------------------
bmi | 7681 28.64962 4.79068 16.06659 58.69388
. table obesity obesity2
| obesity2
obesity | 0 1
----------+-------------
0 | 2,508 169
1 | 222
1,637
. table cohort obesity2 obesity, col row
| obesity and obesity2
| -------- 0 --------
-------- 1 --------
COHORT | 0 1 Total 0 1 Total
----------+-------------------------------------------
0 | 175 1,287 1,462
1 | 2,508 169 2,677 47 350 397
|
Total | 2,508 169
2,677 222
1,637 1,859
22/9/2006
18
Process Steps for WGA Obesity Analysis
Phase 1
Phase 2
Phase 3
Phase 4
Phase 5
SETTINGS OR PROCESS
QC
Single-point
Multipoint
Other
STEPS
Report
QC
Binary trait
Haplotype analysis
Genotyping error
Result
Sensitivity analysis
Statistics
DETAILS
HWE
Population structure
Joint analysis
Pathway analysis
Meta-analysis
Comparisons
OUTCOMES/GOALS
High quality data
Standard
Further information
Further information
Paper
TIME LINE
November?
December?
October?
October?
2007?
