genetic correlations with complete sample overlap

[George]

Hi,

I have two traits with complete sample overlap (N1=5,983,N3=6,089).  The LDsc intercept is low ~1.01 and the sample is very homogeneous and so in estimating the correlations I wanted to constrain the intercept and also add the phenotypic correlation (which I have) to improve the estimate.  However, doing so drastically changes the rg despite the fact that the estimated intercepts are the same as the constrained intercepts.  I was wondering if I need to take the complete sample overlap into consideration when constraining the intercepts, how I do so, and which estimate is more reliable.

Many thanks,

George

If i just estimate it with no extra arguments like in the code below, I get rg=1.1 (0.09)

./ldsc.py \

--rg ~/Desktop/munged_A.txt.sumstats.gz,~/Desktop/munged_B.txt.sumstats.gz \

--ref-ld-chr /Users/g/Downloads/eur_w_ld_chr/ \

--w-ld-chr /Users/g/Downloads/eur_w_ld_chr/ \

--out ~/Desktop/A_B

However if i constrain the intercepts to 1 and add in the phenotypic correlation then it drops to rg=0.61 (0.06)

/ldsc.py \

--rg ~/Desktop/munged_A.txt.sumstats.gz,~/Desktop/munged_B.txt.sumstats.gz \

--intercept-h2 1,1 \

--intercept-gencov 0,0.7 \

--ref-ld-chr /Users/g/Downloads/eur_w_ld_chr/ \

--w-ld-chr /Users/g/Downloads/eur_w_ld_chr/ \

--out ~/Desktop/A_B