Hunting, shooting and fishing...

1
Hunting, shooting and fishing...

• Shooting
• Prior hypothesis
• Hunting
• Promising area
• Fishing
• Angling
• Trawling
• Floundering

2
Hunting, shooting and fishing...

• Shooting
• Prior hypothesis
• Hunting
• Promising area
• Fishing
• Angling
• Trawling
• Floundering

• GE suspected
• Prior publication
• G known, E not
• E (top SNPs)
• E known, G not
• E (GWAS)
• Neither G nor E is known as a risk factor for
  asthma

3
Hunting, shooting and fishing...

• Single test
• plt0.05 or 0.01
• 20-50 tests
• plt0.001 ?
• 550,000 tests
• plt10-6 / 10-7 ?
• 25 million tests
• plt10-9 ?
• plus biology?

• GE suspected
• Prior publication
• G known, E not
• E (top SNPs)
• E known, G not
• E (GWAS)
• Neither G nor E is known as a risk factor for
  asthma

4
Hunting, shooting and fishing...

• 1-stage test
• interaction analysis
• 1-stage test (?)
• interaction analysis
• 2-stage test (?)
• screen then confirm
• 2-stage test
• screen then confirm
• plus biology?

• GE suspected
• Prior publication
• G known, E not
• E (top SNPs)
• E known, G not
• E (GWAS)
• Neither G nor E is known as a risk factor for
  asthma

5
Association of chr17q21 variant rs3894194 with
childhood asthma / wheezy bronchitis
Trend p 2 x 10-7 MAF 44 PARF 27
6
Association of chr17q21 variant rs3894194 with
childhood asthma / wheezy bronchitis
Trend p 2 x 10-7 MAF 44 PARF 27
7
Persons Chromosomes GG Gg gg G G- Cases
n n n D a b Controls n n n D- c d Logistic
regression (G0,1,2) Odds ratio ad/bc Same
results for per-allele odds ratio and significance
8
Formal case-control interaction analysis
(1) Exposed GG Gg gg E G G- Cases n n n D a
b Controls n n n D- c d Unexp. GG Gg gg E- G
G- Cases n n n D e f Controls n n n D- g h In
teraction OR (ad/bc) / (eh/fg) (adfg) / (bceh)
9
Formal case-control interaction analysis
(2) Cases GG Gg gg D G G- Exposed n n n E a
b Unexp. n n n E- e f Controls GG Gg gg D- G G
- Exposed n n n E c d Unexp. n n n E- g h Inte
raction OR (af/be) / (ch/dg) (adfg) / (bceh)
10
Case-only approach to GE interaction
analysis Cases GG Gg gg D G G- Exposed n n n
E a b Unexp. n n n E- e f Interaction OR
(af/be) / 1 Assumes no association between
exposure and genotype in undiseased (Mendelian
randomization). Gains statistical power as no
error in (ch/dg) term. Not statistically
independent of (af/be) / (ch/dg)
11
Two-stage case-control interaction analysis D
D- GG Gg gg All G G- Exposed nn nn nn E ac
bd Unexp. nn nn nn E- eg fh The
screening OR (ac)(fh)/(bd)(eg), an
average of GE associations across cases
controls is statistically independent of the
interaction OR. No assumption of Mendelian
randomization. Gains statistical power in GWAS if
used to select SNPs for formal interaction
testing in 2nd stage at plt0.01.
12
Interaction analysis with multi-level
exposures D D- GG Gg gg All G G- Exposed nn
nn nn E ac bd Unexp. nn nn nn E- eg f
h The screening OR (ac)(fh)/(bd)(eg)
(or the case-only OR in a case-only
design) can also be derived by logistic
regression Modelling exposure as a function of
genotype or Modelling G (0,1) as a function
of exposure (0,1).
13
Interaction with multi-level exposures (step
1) D D- GG Gg gg All G G- High 3 Mediu
m 2 Low 1 None 0 Test for
association of G with E by logistic
regression among cases and controls
combined Modelling G (0,1) as a function of
exposure (0-3).
14
Interaction with multi-level exposures (step
2) Compare association of G with E between cases
and controls, for SNPs with promising screening
ORs Model G (presence of effect allele or
genotype) as a function of level of exposure
(0,1,2...k) in cases Model G (presence of effect
allele or genotype) as a function of level of
exposure (0,1,2...k) in controls Calculate
difference between betas for E1, 2, 3 k within
each study (exposure definition is
consistent). Pool results for these differences
(log interaction ORs) across studies (but
exposure definition or groupings may not be
consistent between studies).
15
Hunting, shooting and fishing...

• 1-stage test
• interaction analysis
• 1-stage test (?)
• interaction analysis
• 2-stage test (?)
• screen then confirm
• 2-stage test
• screen then confirm
• plus biology?

• GE suspected
• Prior publication
• G known, E not
• E (top SNPs)
• E known, G not
• E (GWAS)
• Neither G nor E is known as a risk factor for
  asthma

16
Analytical strategy interactions (1)

• How many asthma-related SNPs should be tested for
  interactions with
• each other (GG interactions)?
• environment and lifestyle factors (GE)?
• Should biological candidate genes be prioritised
  for GE interactions?
• If so, only for biologically plausible
  interaction effects?
• Should interactions be tested only for disease
  outcomes/subgroups that are associated with the
  genotypes?
• At what p value cut-off for the main effect?

17
Analytical strategy interactions (2)

• Should interactions be tested only for the
  genetic model (additive, dominant or recessive)
  that best fits phenotype?
• Probably, to maximise statistical power
• Should interactions be tested in all cohorts
  (with relevant exposure data) even if the
  environmental effect is non-significant in one or
  more cohorts?
• Yes, if we want to meta-analyse eventually
• Should we seek replication of new GE
  interactions before publication?
• Yes, if single study. No, if meta-analysis.

18
Why study interactions?

• Antidote to determinism
• Genetic susceptibility, programming
• Lifecourse approach to disease causation
• More certain identification of causes
• Interaction RR larger than overall RR
• Bias and confounding more easily excluded
• Guidance for public health policy
• Safety for susceptibles

19
(No Transcript)
20
Fat Free- Blotchy wheeler Bloke
Mr Blobby