search

bulik / ldsc

LD Score Regression (LDSC)
GNU General Public License v3.0
652 stars 345 forks source link

Low ratio values - are the genetic correlation results reliable? #203

Open kamiina opened 4 years ago

kamiina commented 4 years ago

Could you please comment: if you have low ratio values (and also low h2 for one of the phenotypes) for two continuous phenotypes (both summary statistics are GC corrected), can the genetic correlation results of these phenotypes be reliable? Or is there some kind of bias, since the intercept is 0.8095 for one of the phenotypes?

Examples: Heritability of phenotype 1

Total Observed scale h2: 0.0473 (0.0254) Lambda GC: 1.0075 Mean Chi^2: 1.0062 Intercept: 0.9866 (0.0082) Ratio < 0 (usually indicates GC correction).

Heritability of phenotype 2/2

Total Observed scale h2: 0.2365 (0.0368) Lambda GC: 1.0195 Mean Chi^2: 1.2092 Intercept: 0.8095 (0.03) Ratio < 0 (usually indicates GC correction)

Thank you!

anbai106 commented 3 years ago

Did you get more intuitions for this?

kamiina commented 3 years ago

Hi, I did not get any reply. I suppose the genetic correlation estimate of these two phenotypes should be quite ok, although the heritability of the first phenotype was estimated to be low (0.0473). LDSC apparently can underestimate the h2, but this should not affect the genetic correlation result. The genetic correlation result for these two phenotypes was: rg=0.8148, se=0.2279, p=0.003. Regarding the low intercepts, I would still like to get some additional insights. Please share your thoughts.