GC and LD score intercept

[Vanessa Oliveira]

Hi,

The Nat Genet paper said the ld score regression intercept is a better correction factor than genomic control. It is not clear for me how to use this information to work with a no bias data set. I am using SCZ summary stats. Should I 1) convert the summary data into ldsc format; 2) use ldsc.py script adding the flag --intercept-h2 N (=1?). Then, how I adjust the p-values?  I am going to use the data set for gene-set analysis.

Sorry if it is a silly question,

Thank you in advance,

Vanessa

[Raymond Walters]

Hi Vanessa,

If I understand correctly your goal is to use LD score regression to get the correction factor that’s analogous to genomic control and use that to adjust your GWAS results?

In that case, you’ll want to:

1) Convert the summary data to ldsc format using munge_sumstats.py (see example in tutorial)

2) Run ldsc.py using --h2 with the basic arguments as described in the tutorial (you don’t want the --intercept-h2 flag)

3) Find the fitted estimate for the intercept in the ldsc output log.

4) Treat the intercept like lambdaGC for adjusting your GWAS results (assuming intercept > 1 so there’s evidence of something to correct). Method depends on what data you have:

- beta (or odds ratio) and SE: multiply the SE by sqrt(intercept)

- z statistics: divide Z stat by sqrt(intercept)

- chi2 statistics: divide chi2 by the intercept

- p-values: recompute after applying above adjustment to test statistics

It should be noted that this kind of correction is still fairly heuristic. The ldsc intercept may still be overly conservative since the intercept may in part reflect model misspecification along with any confounding (see discussion of intercept in this post), and it’s implicitly an adjustment for average inflation genome-wide rather than being tuned to the level of confounding for any particular variant. But despite that is should still outperform classic genomic control, so still reasonable to apply as an alternative.

Hope that helps!

Cheers,

Raymond

--
You received this message because you are subscribed to the Google Groups "ldsc_users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ldsc_users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ldsc_users/090f14a2-e60c-4d3d-a922-8f369608aeff%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.