Genetic variants regulating ORMDL3 expression in asthma: Nature 2007

http://www.nature.com/nature/journal/v448/n7152/full/nature06014.html#online-methods

Association testing

Tests of Hardy–Weinberg equilibrium were performed in cases and controls using the genhw procedure (http://www.biostat-resources.com/stata/) and Stata version 9.2, and SNPs showing Hardy–Weinberg disequilibrium in controls (χ² > 25) were excluded. As the data comprised a mixture of unrelated and related cases and controls, we used logistic regression models with robust sandwich estimation of the variance¹⁵ as implemented in the Stata logit function to model clustering of siblings’ genotypes. Simulations using the MRC-A family structures (data available on request) confirmed that this method appropriately controls the Type I error. Heterogeneity of association between the two main strata (UK and Germany) was assessed by a weighted linear combination test using the results of an additive-effects-only regression analysis within each stratum. X-linked markers were analysed by fitting an additive-effects-only logit model that equates the risks of male hemizygotes with female homozygotes. The TRANSMIT program¹⁶ was used to analyse nuclear family data (including parental genotypes), using the sandwich variance estimation option to robustly incorporate information from multiple affected siblings; confidence intervals for odds ratio estimates were computed as described. The false-discovery rate (FDR) method⁵ was used to assess the overall statistical significance of the genome-wide association results, taking into account the multiple hypothesis testing implications inherent in the analysis of more than 300K SNPs. The FDR thresholds were calculated by applying the QVALUE (http://faculty.washington.edu/%7C[sim]%7Cjstorey/qvalue/) software package¹⁷.

 

Association to transcript abundances

Data from the gene expression experiment were normalized together using the RMA package^{18, 19} to remove any technical or spurious background variation. An inverse normalization transformation step was also applied to each trait to avoid any outliers. Association analysis was applied with Merlin (FASTASSOC option)²⁰. We estimated an additive effect for each SNP and tested its significance using a score test that adjusts for familiality and takes into account uncertainty in the inference of missing genotypes. In the absence of a positive genomic control test, we did not adjust for stratification. We probabilistically inferred missing genotypes²¹ and adjusted for familiality, but not for linkage signal.