Entering edit mode

17 months ago

alessandrotestori7
•
330

Hi. I have the GWAS summary statistics from two different traits. Is there a standard procedure to assign a unique p-value to each SNP in common evaluating its association to both traits? Please let me know; thanks.

You will have to share the summary statistics that you have and which program you used to produce them. I'm imagining PLINK. Also, when you say 'both traits', what do you mean? - is it a multinomial logistics regression model that you are attempting to build?

Edit: If you have statistics for 2 separate traits, then you already have independent P values for each SNP to each trait, including the ones that are common(?).

Hi Kevin! I haven't carried out association analysis myself, I only retrieved these summary statistics (OR, beta, sample size and p-value) from publicly available datasets. If I set a p-value threshold (e.g.: 1E-5) and look for SNPs below this threshold in both datasets, I can find SNPs that are significant in both traits. For example: suppose that rs10 has a p-value of 1E-6 in trait A and a p-value of 1E-7 in trait B; rs20 has a p-value of 1E-9 in trait A and a p-value of 1E-10 in trait B. I conclude that rs10 and rs25 are significantly associated to both traits. I was wondering whether there is a way to combine these p-values (i.e.: combine p-values 1E-5 and 1E-6 from SNP rs10; p-values 1E-7 and 1E-8 from rs25) in order to obtain combined p-values evaluating the association to both traits. Moreover, is it possible to do the same thing considering genomic regions? I mean, suppose rs10 and rs25 were close to each other and thus fall in the same genomic region, is it possible to produce a unique p-value evaluating the association of all SNPs falling in that region to both traits?

Grazie mille. For actually combining P values from independent studies, Fisher's test or the 'sum of P values' method can be used. I direct you to Michael Love's blog post: https://mikelove.wordpress.com/2012/03/12/combining-p-values-fishers-method-sum-of-p-values-binomial/

There are other methods, too. A meta-analysis of the studies combined would be better.

Yes, I would call this linkage disequilibrium. Perhaps this is the direction in which your analysis should proceed.

Great suggestions! Thanks a lot, Kevin.

Prego prego / You're welcome