Post by vvp on Jul 21, 2019 23:30:00 GMT
Dear GCTA community,
I am trying to fit a LMM model of different phenotypes (e.g. log(height)) with two random effects using GCTA with data from a cohort of ~5,000 participants. I used GCTA to compute the GRM and OSCA to compute the methylomic relationship matrix. Now I am trying to use GCTA to fit a LMM with two random effects as described in page 10 of "OSCA: a tool for omic-data-based complex trait analysis" (section: OREML: estimating the proportion of trait variance captured by all DNAm probes). I am using age and sex as fixed effects.
I am interested in estimating the variance components, but the REML estimates I am getting don't make sense to me:
Summary result of REML analysis:
Source Variance SE
V(G1) 0.001238 0.000051
V(G2) 0.000171 0.000042
V(e) 0.000000 0.000051
Vp 0.001409 0.000031
V(G1)/Vp 0.878806 0.023220
V(G2)/Vp 0.121192 0.029093
Sum of V(G)/Vp 0.999998 0.035999
Sampling variance/covariance of the estimates of variance components:
2.574272e-09 2.051217e-10 -1.489689e-09
2.051217e-10 1.733079e-09 -1.661609e-09
-1.489689e-09 -1.661609e-09 2.573096e-09
Source Variance SE
V(G1) 0.001238 0.000051
V(G2) 0.000171 0.000042
V(e) 0.000000 0.000051
Vp 0.001409 0.000031
V(G1)/Vp 0.878806 0.023220
V(G2)/Vp 0.121192 0.029093
Sum of V(G)/Vp 0.999998 0.035999
Sampling variance/covariance of the estimates of variance components:
2.574272e-09 2.051217e-10 -1.489689e-09
2.051217e-10 1.733079e-09 -1.661609e-09
-1.489689e-09 -1.661609e-09 2.573096e-09
So I interpret that the proportion of variance captured by the genetic and methylation components is ~88% and ~12%, respectively, with the error variance component being ~0%. This does not make sense to me as I would not expect the model, or any model, to fit the data so well.
Also, I find that the residual effects printed in the "*.indi.blp" file are smaller (in absolute value) than 5.75995e-07.
Could you provide some help with the interpretation of this output?
Many thanks for your time.
VĂctor.