Negative heritability warning

[Yiling Kuang]

Hi All,

I'm trying to reproduce the results for User Specified Models with SNP Effects. I found the round 1 Neale Lab UK Biobank GWAS results are not available now, so I tried the round 2 Neale Lab UK Biobank GWAS results.

When I run the ldsc function, there is a warning:

In ldsc(traits = traits, sample.prev = sample.prev, population.prev = population.prev, : Your genetic covariance matrix includes traits estimated to have a negative heritability.

And when I run the userGWAS function, there is an error:

Error in { : task 1 failed - "system is computationally singular: reciprocal condition number = 4.57139e-17"

I think according to the definition of SNP heritability it should be positive?  I wonder if anyone else has encountered this issue and has some tips for that?

Thanks for your help.

Best,

Kylie

[Genomic SEM Users]

Hi,

since LDSC estimates heritability as the slope of a regression of the Z2 on the LD score, that slope can become negative. This cna happen with sumstats derived from certain tpes of mixed modesl (not likely the case here) or for low powered GWAS (so low N or low N_cases)/

Hope this helps!

Michel