Replication in genome-wide association studies

Abstract

Replication helps ensure that a genotype-phenotype association observed in a genome-wide association (GWA) study represents a credible association and is not a chance finding or an artifact due to uncontrolled biases. We discuss prerequisites for exact replication; issues of heterogeneity; advantages and disadvantages of different methods of data synthesis across multiple studies; frequentist vs. Bayesian inferences for replication; and challenges that arise from multi-team collaborations. While consistent replication can greatly improve the credibility of a genotype-phenotype association, it may not eliminate spurious associations due to biases shared by many studies. Conversely, lack of replication in well-powered follow-up studies usually invalidates the initially proposed association, although occasionally it may point to differences in linkage disequilibrium or effect modifiers across studies.

Figure 1

The relationship between the Bayes Factor and p-value for different sample sizes and minor allele frequencies (left panel: minor allele frequency of 40%; right panel: 5%). The dashed line represents the Bayes Factor necessary to achieve posterior odds in favor of association of 3:1 or greater, assuming the prior odds of association are 1:99,999. Bayes Factors were calculated for a case-control study with equal numbers of cases and controls, assuming the expected value of the absolute value of the log odds ratio is log(1.15), and assuming a “spike and smear” prior. Calculations use equations (4) and (5) from Ioannidis (2008),[13] with σ² = I_ββ⁻¹ − I_αβ⁻¹ [I_αα⁻¹]⁻¹I_βα⁻¹, where I is the Fisher information from simple logistic regression log(odds) = α + β G_additive calculated under the null (β=0).

Figure 2

Quantile-quantile plots for fixed-effect and random-effect meta-analyses of the 13 studies in the initial PanScan genome-wide association study of pancreatic cancer. The genomic control inflation factors λ_GC for the fixed-effect and random effect analyses were 0.84 and 1.00, respectively. λ_GC was calculated as the median observed chi-squared test statistic divided by the median of a chi-squared distribution with one degree of freedom.