Should we check if h2 from sum stats via LDSC are the same/similar to h2 from HE regression/GREML?
Replying to “Should we check if h2 from sum stats via LDSC are …”:
If you have access to suitable data to check that, you can.
But one of the advantages of LDSC is that you can do it with sumstats only (no requirement for individual-level data), so often people using them don’t have individual-level data where they could do HE or GREML
Replying to “Should we check if h2 from sum stats via LDSC are …”:
One would typically expect LDSC h2 to be a little lower than HE/GREML h2 for a trait, because HE and GREML typically incorporate other forms of genetic variation beyond just SNPs
Replying to “Should we check if h2 from sum stats via LDSC are …”:
GCTA/GREML h2 are usually higher than LDScore - LDscore is based on snps and things in LD, GCTA/GREML is based on relatedness which collects more along the way
Replying to “Should we check if h2 from sum stats via LDSC are …”:
On Ted and Sarah’s point, this paper is a good read on the assumptions of different methods and impact on the h2 estimates: Comparison of methods that use whole genome data to estimate the heritability and genetic architecture of complex traits - PubMed
Replying to “Should we check if h2 from sum stats via LDSC are …”:
Evans LM, Tahmasbi R, Vrieze SI, Abecasis GR, Das S, Gazal S, Bjelland DW, de Candia TR; Haplotype Reference Consortium; Goddard ME, Neale BM, Yang J, Visscher PM, Keller MC. Comparison of methods that use whole genome data to estimate the heritability and genetic architecture of complex traits. Nat Genet. 2018 May;50(5):737-745. doi: 10.1038/s41588-018-0108-x. Epub 2018 Apr 26. PMID: 29700474; PMCID: PMC5934350.