Genetic Association Studies: Analysis Strategy Considerations

1
Genetic Association Studies Analysis Strategy
Considerations
Eleftheria Zeggini elez_at_well.ox.ac.uk
2
Overview

• Quality control
• -single nucleotide polymorphism genotyping
  quality
• -Hardy-Weinberg equilibrium
• -frequency comparisons
• -linkage disequilibrium
• Analysis
• -single-point
• -haplotype-based
• Interpretation
• -adjusting for multiple testing
• -replication
• -meta-analysis

3
Quality Control Genotyping

• Duplicate genotype consistency
• check duplicates across plates
• flag / drop SNPs with inconsistencies
• Genotype call quality scores
• set threshold for acceptable quality
• plot quality scores v. physical distance
• Genotyping completeness
• proportion of missing data

4
QC Hardy-Weinberg equilibrium

• Check HWE in cases and controls separately
• Control genotypes should be in HWE
• Deviation from HWE may indicate
• problematic genotyping
• population stratification
• biological mechanism

5
QC HWE in cases

• Deviation from HWE in cases may indicate
• problematic genotyping
• population stratification
• biological mechanism
• underlying disease association
• Will affect downstream choice of analysis
  strategy
• Allele and haplotype-specific tests not
  applicable due to non-independent parental
  allele transmission

6
QC Evaluating deviation from HWE

• Check genotype distribution
• Excess heterozygotes
• Excess homozygotes
• implies recessive model
• can approximate expected effect size
• e.g. ls GRR
• Plot HWE p values v. physical distance

7
QC Evaluating deviation from HWE
8
QC Genotype frequency comparison

• Review literature for studies including SNPs
  tested
• Compare SNP frequencies with data from same
  population / populations of similar ancestry
• sample size
• genotyping method
• Flag SNPs displaying gross inconsistencies

9
QC Linkage disequilibrium patterns

• Characterise pairwise LD patterns in cases and
  controls separately
• Pairs of SNPs separated by long physical distance
  are not likely to be in strong LD
• Correlate strong apparent long-range LD with
• minor allele frequency
• genotype quality scores
• proportion of missing data

10
QC Wish list

• High quality genotyping
• Low proportion of missing data
• Controls and cases in HWE
• or
• Deviation from HWE attributable to plausible
  association
• Similar genotype frequencies with public domain
  data
• No unexplainable strong apparent long-range LD

11
Analysis

• Single-point
• Haplotype-based

12
Analysis Single-point

• Compare individual SNP frequencies between cases
  and controls
• 5 different models can be tested
• General, dominant, recessive, multiplicative,
  additive

13
Analysis Which model?

• General Compare genotype distributions
  AA, AB, BB between cases and controls
• Dominant (AA AB) v. BB
• Recessive AA v. (AB BB)
• Multiplicative A v. B
• Additive AA AB BB

14
Analysis Fish and tell

• No prior reason to expect specific model
• Good practice to
• report number of models tested
• report sub-analysis results
• adjust for multiple testing
• replicate positive findings

15
Association testing

• Direct functional variant affecting disease risk
  is tested
• Indirect SNPs in LD with the true functional
  variant are tested

16
Indirect association testing

• Relies on allelic association between tested SNPs
  and functional variant
• Influenced by effect size, disease variant
  frequency, marker allele frequency, extent of LD
• Increase power to capture disease association by
  matching marker and disease variant allele
  frequencies

17
Analysis Our friend the haplotype

• Haplotypes can contain more information than
  single markers
• If the disease variant is not directly tested,
  haplotypes can help increase power by matching
  variant allele frequency
• Haplotypes can indicate synergistic effects

18
Analysis Estimating haplotypes

• Drawbacks of haplotype inference in unrelated
  individuals methodological assumptions, missing
  data, uncertainty
• Loss of information compared with genotype-based
  analysis approaches when uncertainty not
  incorporated

19
Analysis Haplotype blocks v. moving haplotype
window

• Selection of haplotype-based analysis strategy
• Characterise blocks and compare frequencies of
  within-block haplotypes
• Compare frequencies of moving haplotype window
  across region of interest
• Consequences for downstream interpretation of
  results

20
Analysis Single-point v. haplotype-based

• Relative efficiency of approaches depends on
  tested SNP density
• Haplotype tagging SNPs
• Majority of haplotype information encapsulated by
  htSNPs
• Complementary approaches

21
Analysis Wish list

• Single-point analyses point to one specific
  variant / gene / subregion
• Haplotype analyses confirm finding and help
  narrow down the disease gene-containing interval
• Unfortunately, the wish list scenario rarely
  occurs multiple associations, hard to discern
  true positives

22
Will the real disease gene please stand up?

• Adjusting for multiple testing
• Understanding LD and non-independence of markers
• Replication
• Meta-analysis

23
Interpretation Adjusting for multiple testing

• Guarding against false positives
• Family-wise error rate (FWER) v. false discovery
  rate (FDR)
• false positive rate
• of false positives / of true null tests
• false discovery rate
• of false positives / of significant tests

24
Interpretation Non-independence of markers

• Important to characterise the underlying
  structure of LD
• Markers in variable levels of LD with the true
  disease variant will pick up the same association
  signal
• Conditional tests can help dissect out true
  positives

25
Interpretation Replication and meta-analysis

• Important to achieve independent replication of
  positive findings
• Inconsistency of published results variable
  effect sizes, power, extent of LD
• Meta-analysis, combining many different studies,
  can provide better estimates of effect sizes

26
Conclusion What makes a good association study?

• Meticulous study design
• Accurate genotyping
• Stringent QC
• Carefully selected analysis plan
• Review of complementary evidence
• Replication replication replication