Post by anamaria on Sept 4, 2020 0:34:09 GMT
Hello,
I am not sure if this is a good thing to do but I was trying to assess genetic correlation using bivariate GREML as implemented in the GCTA on a single set of subjects. In detail I have some old imputed data and I do have a new imputed data on the same subjects so I was running bivariate analysis where in pheno file I have one trait case/control designation for old data and another traint case/control designation for new data.
My pheno file looks like this:
> head(a)
FID IID new old
1 0 fam0110_G110 2 2
2 0 fam0113_G113 2 2
3 0 fam0114_G114 2 2
4 0 fam0117_G117 2 2
5 0 fam0119_G119 1 1
7 0 fam0127_G127 2 2
...
And I was running this:
./gcta64 --reml-bivar --grm Merge --pheno combined_old_new --out GCTA_trait2
I got:
Reading IDs of the GRM from [Merge.grm.id].
1634 IDs read from [Merge.grm.id].
Reading the GRM from [Merge.grm.bin].
GRM for 1634 individuals are included from [Merge.grm.bin].
Reading phenotypes from [combined_old_new].
There are 2 traits specified in the file [combined_old_new].
Traits 1 and 2 are included in the bivariate analysis.
Non-missing phenotypes of 1536 individuals are included from [combined_old_new].
1535 individuals are in common in these files.
1525 non-missing phenotypes for trait #1 and 1533 for trait #2
768 cases and 757 controls for trait #1
778 cases and 755 controls for trait #2
Note: we can specify the disease prevalence by the option --reml-bivar-prevalence so that GCTA can transform the variance explained to the underlying liability scale.
Constructing the covariances
Performing bivariate REML analysis ... (Note: may take hours depending on sample size).
3058 observations, 2 fixed effect(s), and 6 variance component(s)(including residual variance).
Calculating prior values of variance components by EM-REML ...
Updated prior values: 0.0978423 0.0978288 0.0795533 0.0972487 0.0972211 0.0802486
logL: 1310.13
Running AI-REML algorithm ...
Iter. logL V(G)_tr1 V(G)_tr2 C(G)_tr12 V(e)_tr1 V(e)_tr2 C(e)_tr12
Note: to constrain the correlation being from -1 to 1, a genetic (or residual) variance-covariance matrix is bended to be positive definite. In this case, the SE is unreliable.
1 2257.59 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 (2 component(s) constrained)
Error: the variance-covaraince matrix V is not invertible.
An error occurs, please check the options or data
Is it not possible to run this analysis if the cases and controls are the same?
What else I can do to calculate genetic correlation? I don't have GWAS summary stats (if I do I would do LDSC) I only have genotype data.
Please advise,
My pheno file looks like this:
> head(a)
FID IID new old
1 0 fam0110_G110 2 2
2 0 fam0113_G113 2 2
3 0 fam0114_G114 2 2
4 0 fam0117_G117 2 2
5 0 fam0119_G119 1 1
7 0 fam0127_G127 2 2
...
And I was running this:
./gcta64 --reml-bivar --grm Merge --pheno combined_old_new --out GCTA_trait2
I got:
Reading IDs of the GRM from [Merge.grm.id].
1634 IDs read from [Merge.grm.id].
Reading the GRM from [Merge.grm.bin].
GRM for 1634 individuals are included from [Merge.grm.bin].
Reading phenotypes from [combined_old_new].
There are 2 traits specified in the file [combined_old_new].
Traits 1 and 2 are included in the bivariate analysis.
Non-missing phenotypes of 1536 individuals are included from [combined_old_new].
1535 individuals are in common in these files.
1525 non-missing phenotypes for trait #1 and 1533 for trait #2
768 cases and 757 controls for trait #1
778 cases and 755 controls for trait #2
Note: we can specify the disease prevalence by the option --reml-bivar-prevalence so that GCTA can transform the variance explained to the underlying liability scale.
Constructing the covariances
Performing bivariate REML analysis ... (Note: may take hours depending on sample size).
3058 observations, 2 fixed effect(s), and 6 variance component(s)(including residual variance).
Calculating prior values of variance components by EM-REML ...
Updated prior values: 0.0978423 0.0978288 0.0795533 0.0972487 0.0972211 0.0802486
logL: 1310.13
Running AI-REML algorithm ...
Iter. logL V(G)_tr1 V(G)_tr2 C(G)_tr12 V(e)_tr1 V(e)_tr2 C(e)_tr12
Note: to constrain the correlation being from -1 to 1, a genetic (or residual) variance-covariance matrix is bended to be positive definite. In this case, the SE is unreliable.
1 2257.59 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 (2 component(s) constrained)
Error: the variance-covaraince matrix V is not invertible.
An error occurs, please check the options or data
Is it not possible to run this analysis if the cases and controls are the same?
What else I can do to calculate genetic correlation? I don't have GWAS summary stats (if I do I would do LDSC) I only have genotype data.
Please advise,