Does it makes sense to do some partitioned heritability analysis when h2 is 0.008(0.010) ?

[Veera Manikandan]

Hi Raymond and LDSC team, 

I am running a GWAS on a secondary phenotype in iPSYCH (cases = 1300 controls = 65000)

I got h2 = 0.008(0.010)

I just tried running some partitioned h2 analysis, and I got some significant enrichments (p values are in range of 0.01 to 0.05, but enrichments numbers pr(h2)/pr(SNPs) are large like 100, 40,60). If these estimates make sense, given very low h2 of 0.008 ?

I would like your opinion on this. 

Thanks

Veera 

[Raymond Walters]

Hi Veera,

I’d be very skeptical of those enrichment results for a couple reasons:

1) h2=.008 with se=.010 is non-significant. It’s hard to say a particular annotation is enriched for heritability if there’s not overall heritability. The original partitioned heritability paper restricted to a minimum z score of 7 (p=1e-12) for heritability before attempting partitioning. We tend to be somewhat more lenient then that, but partitioning non-significant heritability certainly isn’t recommended.

2) Nominal significance of p=.01-.05 isn’t terribly surprising given multiple testing of e.g. 24 main annotations in the baseline model. With just those main annotations (i.e. ignoring their border regions) that means a bonferroni-corrected significance threshold of p=.002. 

3) Since the enrichment calculation involves the proportion of heritability it requires division by the total heritability, and h2=0.008 is a very small denominator to have in that calculation. The resulting estimate is generally going to be very unstable, which explains your very large enrichment values; presumably those large estimates also have large SEs that reflect this instability.

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/76bf8fa9-4539-46a2-9c0c-14b3025668b6%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Veera Manikandan]

Hi Raymond,

Thank you very much for the info. That makes sense. 

Veera