I am planning to conduct meta analysis of a relatively rare phenotypes. I have come across studies from 1980s and 1990s where authors studied one of two loci. There are some recent publication with GWAS results. Can I combine results from these two methodologies. Can anyone suggest publications and software to guide me on this.
I'd be worried less about heterogeneity (that can be accounted for) and more about a number of other problems that have plagued old-fashoned candidate-gene studies, resulting in thousands of spurious reports of association.
First, there is publication bias. You don't collect 200 or 300 cases and as many controls (the typical sample size in pre-2007 reports) to do just one SNP in just one canadidate gene. But, typically, these studies only publish the one out of 20 SNPs that would have p
One has to be concerned about heterogeneity. Typically GWAS genotyping provides far more uniformity in quality than individual site genotyping, especially if it was from so many years ago due to a much higher error rate (technique related) in the earlier genotyping.
Aside the issue of older technology you can combine data from different technologies using most meta-analysis software, such as GWAMA which also has metrics for heterogeneity. For example in GWAS or GWAS meta-analyses studies authors will follow-up a few variants for de novo replication and type them in a plex like Sequenom. Summary statistics from each study can then be combined in a meta-analysis framework using fixed-effects or random-effects models. The latter is recommended when there is significant heterogeneity but it is not a powerful model so it is usually avoided.
In a case/control analysis you will need to know the OR and CIs whereas for continuous phenotypes you would need to know the beta and SE (as well as effect allele). For continuous phenotypes though you need to ensure that your phenotypes have undergone the same phenotype transformation (if at all) as the reported trait associations so that the betas are compatible.
I'd be worried less about heterogeneity (that can be accounted for) and more about a number of other problems that have plagued old-fashoned candidate-gene studies, resulting in thousands of spurious reports of association.
First, there is publication bias. You don't collect 200 or 300 cases and as many controls (the typical sample size in pre-2007 reports) to do just one SNP in just one canadidate gene. But, typically, these studies only publish the one out of 20 SNPs that would have p
- Badges
- Science topic
- Similar topics
- Biological Science
- Genetics