search

privefl / bigsnpr

R package for the analysis of massive SNP arrays.
https://privefl.github.io/bigsnpr/
183 stars 43 forks source link

LDSC heritability #425

Closed lingjoyo closed 1 year ago

lingjoyo commented 1 year ago

Hallo Florian

I'm trying to calculate the heritability of coronary artery disease (CAD) which is reported to be around 40%. I used the GWAS data which was also used in your paper. But the heritability calculated using snp_ldsc() is only ~9%. Do you think it's reasonable?

privefl commented 1 year ago

It depends. How many variants do you use? Do you talk about the same kind of heritability (full additive vs SNP heritability)? Are these estimates on the same scale (observed vs liability scale)?

lingjoyo commented 1 year ago

Hi Florian,

Thanks for your instant reply.

I used 1.2 million SNPs that existed in HapMap3+.Then I randomly select 10000 samples from UK Biobank to calculate the LD correlation matrix. The example data used in the instruction https://privefl.github.io/bigsnpr/articles/LDpred2.html only has ~45k variants. But the heritability reached 0.298.

What do the full additive and SNP heritability mean respectively?

privefl commented 1 year ago

The numbers you used are fine. The example data is simulated.

Have a look at the literature for the difference between the SNP heritability and the narrow-sense (additive) heritability. Or maybe at this lecture: https://www.youtube.com/watch?v=kG-Zn6Wq6Ac.

privefl commented 1 year ago

Any update on this?