Greetings everyone,
I performed an association analysis of mitochondrial variants to a binary phenotype on circa 9000 samples, of which 6700 were cases and 2300 were control samples. I used 20 principal components and mitochondrial copy number (obtained through the MitoHPC pipeline) as covariates. I know that for nuclear GWAS studies, a pvalue threshold of 5 × 10^-8 is widely used to identify significant variants. What about mitochondrial variants? In a first istance, I just used 0.05/n°analyzedVariants, wich results in a range of 10^-6, but it also resulted in very few significant variants, so maybe it is too stringent. What is your opinion on this, how would you evaluate the threshold? For the analysis I resorted to PLINK2 with the blueprint command line:
./plink2 --pfile /path/to/allSamples --glm --covar /path/to/allSamples.covar --covar-variance-standardize --out /path/to/out/allSamples
Thank you in advance
Hello,
0.05/n°analyzedVariants corresponds to the bonferroni correction, which corrects for multiple testing, but is stringent.
Maybe you could have a look at the FDR correction (False Discovery Rate), which only looks at significant results.
You can also make a QQplot (your observed p-values, vs the theoritical p-values, to look for inflation/deflation of the distribution of your p-values).
It is also possible that only few mitochondrial variants are strongly associated with the phenotype.
Alternatively, you can perform rare variants association tests (SKAT, burden tests....) to look for rare variants instead of common variants.
20 principal components seems a lot as covariates, but that is just my opinion.