I am not sure what the SNP LD contributes to interpretation of shared genetics across traits; If a SNP is highly significant in sumstats for both trait 1 and trait 2, would LD score regression “devalue” this SNP, in terms of correlation…? What would the interpretation be?
Replying to “I am not sure what the SNP LD contributes to inter…”:
It would devalue it only relative to its amount of LD, meaning that we appropriately account for it being in a big (or small) LD block — every SNP in an LD block of that same size would be ‘rescaled’ in a similar fashion, so a super-significant SNP would still stick out relative to its peers (but now making it a fair comparison to smaller/larger LD blocks)
Replying to “I am not sure what the SNP LD contributes to inter…”:
As a default LDSC removes very highly significant SNPs if I recall correctly chi2 > 30 SNPs are removed because as you imagine it could destabilise the genome-wide calculation. For example, with the APOE locus in Alzheimer’s disease or FTO with BMI
Replying to “I am not sure what the SNP LD contributes to inter…”:
what exactly is rescaled according to the LD score? I am a bit uncertain what the output of LDSC actually is, but it appears that being in a High as opposed to Low LD block is “Better” or more indicative of shared SNPs across traits…?
Replying to “I am not sure what the SNP LD contributes to inter…”:
The effect size (chi-square for a single trait, or cross-product of the association for correlations) is rescaled — the major assumption here is that because traits are highly polygenic across the whole genome, SNPs in high LD are just going to have systematically bigger signals, and that’s due to LD not actual effects more frequently residing in those parts of the genome. If you don’t correct for this somehow, your results are mega confounded by LD!
Replying to “I am not sure what the SNP LD contributes to inter…”:
So does LDSC serve as an adjustment of existing GWAS, a seperate analysis, or both? I understand the idea of adjusting signals by LD, but in terms of discerning shared genetics across traits, I got the idea that shared SNPs with High LD were considered more evident of shared genetics than a similar effect SNP with low LD…? But maybe I misunderstand.
Replying to “I am not sure what the SNP LD contributes to inter…”:
it is a separate analysis to test specific hypotheses
Replying to “I am not sure what the SNP LD contributes to inter…”:
The amount of LD is independent of its (true) genetic signal — hence why we need to use LDSC to disambiguate the two