|
|
GCTA `GREML` estimates were fitted as follows:
|
|
|
|
|
|
```{r eval = FALSE}
|
|
|
gcta64 --reml-bivar 1 2 --grm-gz large.gcta --pheno large.gcta.phe
|
|
|
--reml-bivar-lrt-rg 0 --thread-num 4 --out Y1_Y2 --reml-maxit 1000 >
|
|
|
large.gcta.Y1Y2.log`
|
|
|
```
|
|
|
|
|
|
which yielded the estimates:
|
|
|
|
|
|
```{r eval = FALSE}
|
|
|
Summary result of REML analysis (shortened output):
|
|
|
Source Variance SE
|
|
|
V(G)/Vp_tr1 0.361387 0.020470
|
|
|
V(G)/Vp_tr2 0.635426 0.015053
|
|
|
rG 0.557367 0.026039
|
|
|
```
|
|
|
|
|
|
Similar estimates can be obtained with `grmsem`, fitting a Cholesky model:
|
|
|
|
|
|
```{r eval = FALSE}
|
|
|
load("ph.large.RData")
|
|
|
load("G.large.RData")
|
|
|
ph.biv<-ph.large[,c(1,2)]
|
|
|
fit <- grmsem.fit(ph.biv, G.large, LogL = TRUE, estSE = TRUE)
|
|
|
print(fit)
|
|
|
st.fit <- grmsem.stpar(fit)
|
|
|
print(st.fit)
|
|
|
st.var.fit <- grmsem.var(st.fit)
|
|
|
print(st.var.fit)
|
|
|
```
|
|
|
|
|
|
The standardised variances are (shortened output):
|
|
|
|
|
|
```{r eval = FALSE}
|
|
|
$VA
|
|
|
1 2
|
|
|
Y1 0.3613708 0.2669526
|
|
|
Y2 0.2669526 0.6349136
|
|
|
|
|
|
$VA.se
|
|
|
1 2
|
|
|
Y1 0.01943807 0.01555813
|
|
|
Y2 0.01555813 0.01355490
|
|
|
|
|
|
$RG
|
|
|
1 2
|
|
|
Y1 1.0000000 0.5573145
|
|
|
Y2 0.5573145 1.0000000
|
|
|
|
|
|
$RG.se
|
|
|
1 2
|
|
|
Y1 7.179876e-18 0.0239077
|
|
|
Y2 2.390770e-02 0.0000000
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
|
Both, GREML and GRMSEM estimates from bivariate models, correspond to the GRMSEM estimates from the quad-variate Cholesky model (see example). |
|
|
\ No newline at end of file |