SNP Association p-values

[Zachary Tatom]

Hi,

I have a DO manuscript that was just reviewed and one of the reviewers specifically requested p-values for SNP associations (or at least a p-value threshold rather than a 1.5-LOD drop).

Is this possible in R/QTL2? I understand some convert LOD scores directly to p-values because both are based on an LRS (LOD = `LRS/log10(2)`, p = `chisq(LRS, df = 2)`), but is that a statistically valid approach?

Otherwise, how could we go about getting a valid p-value or p-value threshold for our SNP associations?

Thanks,

Zack

[Fred Boehm]

Hi, Zack,

i'm no expert, but I have seen many researchers use permutation tests for this task. qtl2, i think, contains a function called "scan1perm" that can perform the permutation scans. Here is a link to some of Karl's documentation: https://kbroman.org/qtl2/assets/vignettes/user_guide.html#Performing_a_permutation_test

Please note that he's also written a `summary` method for the output so that you get quantiles of the max lods (ie, the max lods for each genome-wide scan of permuted data), ie, with R code like:

summary(operm, alpha=c(0.2, 0.05))

for more about permutation tests in this setting, I've referred in the past to: 

https://academic.oup.com/genetics/article/142/1/285/6016633

https://academic.oup.com/genetics/article/211/2/495/5931111

I might be mistaken, but i think that karl & saunak's 2009 book on qtl mapping (with old qtl r package) might also discuss permutation tests in this setting. https://link.springer.com/book/10.1007/978-0-387-92125-9

I hope that this is useful and that other readers might correct my message if I'm mistaken.

Cheers,

fred

-----------------

Fred Boehm

https://fboehm.us/

--
You received this message because you are subscribed to the Google Groups "R/qtl2 discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rqtl2-disc+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl2-disc/73c56fcd-3019-4ba9-853e-3aad690903c3n%40googlegroups.com.

[Karl Broman]

Hi, Zack,

I would go ahead and calculate the p-values as you proposed, assuming a chi-square null distribution. No need to worry about “statistically valid”, as it’s just a monotonic transformation. 

Personally, I don’t see how SNP p-values or p-value thresholds have any relation at all to the 1.5-LOD drop, but I generally follow the rule, “give the reviewers what they want, as long as it’s not complete nonsense”.

I agree with Fred that, to get a threshold on the SNP p-values, it’d be best to use a permutation test. But we haven’t yet implemented a permutation test for the scan1snps() results, so you’d need to roll your own.

karl

[Zachary Tatom]

Thank you for your reply. I'm aware of the permutation function for overall QTL p-values (the results from `scan1` + `scan1perm`), but I'm asking specifically about SNP associations (the results from `scan1snps`).