Seeking to make Bonferoni corrections. What does number of tests mean when I check the association of a single SNP. Do I have to apply 3 since I have three parameters AA, Aa and aa) or 2 since I analyse dominant and recessive models of association (AA vs Aa+aa etc.).
Just 2 if you want to apply it....2 refers to the model of association (dominant and recessive).
How many SNPs and traits do you have? You rarely use Bonferonni for less than 5 tests. You should do the number of tests run which is nSNP * ntrait *nmodels run, assuming all models on all SNPs for all traits. However, I doubt your tests are truly independent so a Bonferonni may be too stringent - consider FDR.
I have 5 SNPs but they are in a single gene. I have checked the association with the disease and with several clinical parameters. Since the cohort is small the referee wants me to perform Bonferonni correction. My problem was how to chose to number of comparisons - 5 SNPs x 10 groups etc. I have spent two days reading articles and I have not found any steady paragon.
Tnank you!
You should correct according to linkage disequilibrium among the SNPs...try with Nyholt method (http://gump.qimr.edu.au/general/daleN/SNPSpD/ ) or just according the LD blocks
As you are working on revisions, I advise you to keep the correction very conventional and straightforward. This will stop a whole discussion and just get your paper out there. The conventional Bonferroni correction is the alpha / the no. of tests. This will be your new p-value for significance. Assuming you want an alpha of .05, and assuming 5 SNPs and 1 disease, and you ran one model for each SNP (I.e. just one test of association), then your new p-value will be .05/5 = .025. If you ran dominant AND recessive (why did you do this? You should have just picked a model based on the trait mean by the genotype, to save yourself a costly correction). Then you ran 10 tests: 5 SNPs * 2 models each. Your new p-value is .05/10 = .005.
For just 5 SNPs, within a gene, I would not worry about LD. In fact, as an editorial board member in that scenario I usually think authors are trying to sneak something past that would not survive a stringent correction.
Trust me, when reviewers ask for a Bonferroni, this is exactly what they want to see. If they ask for a 'correction for multiple testing' but do not state what, then you have more leeway.
- Badges
- Science topic
- Similar topics
- Publishing