Computational protocol: Meta-analysis indicates that common variants at the DISC1 locus are not associated with schizophrenia

Similar protocols

Protocol publication

[…] One of the problems encountered in previous attempts at meta-analysis is the lack of overlap between studies in terms of tested SNPs. We addressed this problem by imputing genotypes for studies where we had full genotype data. Imputation was carried out using IMPUTE v.2,, with reference panels of CEU haplotypes from HapMap 3 and the 1000 Genomes pilot 1. The CEU haplotypes are taken from samples from the Centre d'Etude du Polymorphisme Humain collected from individuals living in Utah, USA, with ancestry from northern and western Europe. The haplotype data were obtained from the IMPUTE website (http://mathgen.stats.ox.ac.uk/impute/impute.html). For studies where we did not have full genotype data (we had genotype counts or MAFs), we imputed MAFs using the linear predictor implemented in the BLIMP package and the CEU haplotypes from 1000 Genomes pilot 1. We attempted to impute for all studies that genotyped at more than 10 SNPs. [...] Log-ORs (equivalent to logistic regression coefficients) and their standard deviations were calculated for each SNP in each study and then combined across studies using fixed and random effects models. We tested whether any of the pooled P-values were more significant than the thresholds calculated above. We also tested whether the distribution of P-values differed significantly from that expected under the null hypothesis. For each SNP, we investigated whether there was any evidence for heterogeneity using the I2 statistic.For the imputed genotypes, we calculated the test statistics using SNPTEST, which takes into account the genotype uncertainty generated by the imputation. For imputed allele frequencies, we calculated the proportion of the genotypic variance explained by the imputation, and filtered out any SNP where this was below 80%. We then calculated test statistics as for the original data. These were then combined with the test statistics from genotyped SNPs as above. We considered separately: (1) the genotyped data; (2) the genotyped data plus the imputed data, but considered only loci that were genotyped in at least one study; and (3) the genotyped data plus all the imputed data. We removed each study from the meta-analysis in turn to determine whether any single study had a disproportionate effect on the analysis. We also calculated Wakefield approximate Bayes factors from estimates of the standardized effect sizes and their variances, assuming a prior for the effect size that was normally distributed with mean 0 and variance 0.015. This prior is equivalent to the belief that for each SNP, there is a 95% probability that its OR lies between 0.79 and 1.28. […]

Pipeline specifications

Software tools IMPUTE, BLIMP, SNPTEST
Application GWAS
Chemicals Nucleotides