|
Post by roberto on Nov 13, 2013 14:19:15 GMT
Hello,
some questions concerning the bivariate analyses:
1. There are two values in the genetic correlation row (rG). The first is rG, of course, but what is the 2nd value? Is it a standard error? a P-value?
2. Is there a way of using the LRT value to assess the reliability of a rG estimate?
3. Is it expected that rG for a phenotype with itself be rg=NA instead of rg=1?
thank you in advance!
best, roberto
|
|
|
Post by Jian Yang on Nov 14, 2013 13:47:09 GMT
1) It's the SE 2) Yes, --reml-bivar-lrt-rg 3) rg = 1
|
|
|
Post by roberto on Nov 14, 2013 16:54:21 GMT
thank you Jian!
and for (2) I should use --reml-bivar-lrt 0, right?
|
|
|
Post by Jian Yang on Nov 17, 2013 13:01:20 GMT
--reml-bivar-lrt-rg 0 to test against rg = 0 --reml-bivar-lrt-rg 1 to test against rg = 1 --reml-bivar-lrt-rg -1 to test against rg = -1
|
|
|
Post by roberto on Nov 19, 2013 13:13:58 GMT
Hi Jian,
The reason I was asking what should be the correlation of a phenotype with itself is that I have two phenotypes very correlated, and likely to be very genetically correlated as well (head size and brain volume).
When I estimate their rG I get 0.96±0.04. However, if I try to test this rG against 1 or against 0 (or anything else), the execution fails with an "Error: the information matrix is not invertible." (I pasted the log at the end). To check if a too strong rG could make the algorithm fail I tried computing rG of the same phenotype twice (head size vs head size), and it did indeed produce the same "matrix not invertible" error. Is there any alternative to test if the rG I'm getting is significantly >0?
thanks in advance!
roberto
******************************************************************* * Genome-wide Complex Trait Analysis (GCTA) * version 1.21 * (C) 2010-2013 Jian Yang, Hong Lee, Michael Goddard and Peter Visscher * The University of Queensland * MIT License ******************************************************************* Analysis started: Tue Nov 19 14:08:13 2013
Options: --grm-bin rto/data/imagen2/results/grm-all/grm-all --keep rto/data/imagen2/data/phenotype/keep.txt --reml-maxit 200 --reml-bivar-lrt-rg 0 --reml-bivar 1 2 --pheno pheno.txt --qcovar qcovar.txt --covar covar.txt
Note: This is a multi-thread program. You could specify the number of threads by the --thread-num option to speed up the computation if there are multiple processors in your machine.
Reading IDs of the GRM from [rto/data/imagen2/results/grm-all/grm-all.grm.id]. 2087 IDs read from [rto/data/imagen2/results/grm-all/grm-all.grm.id]. Reading the GRM from [rto/data/imagen2/results/grm-all/grm-all.grm.bin]. Reading the number of SNPs for the GRM from [rto/data/imagen2/results/grm-all/grm-all.grm.N.bin]. Pairwise genetic relationships between 2087 individuals are included from [rto/data/imagen2/results/grm-all/grm-all.grm.bin]. Reading phenotypes from [pheno.txt]. There are 2 traits specified in the file [pheno.txt]. Traits 1 and 2 are included in the bivariate analysis. Nonmissing phenotypes of 2085 individuals are included from [pheno.txt]. Reading quantative covariates from [qcovar.txt]. 11 quantative covarites of 1802 individuals read from [qcovar.txt]. Reading discrete covariate(s) from [covar.txt]. 2 discrete covariate(s) of 2089 individuals are included from [covar.txt]. 1765 individuals are kept from [rto/data/imagen2/data/phenotype/keep.txt]. 1765 individuals are in common in these files. 1765 non-missing phenotypes for trait #1 and 1765 for trait #2 11 quantitative variable(s) included as covariate(s). 2 discrete variable(s) included as covariate(s).
Performing bivariate REML analysis ... (Note: may take hours depending on sample size). 3530 observations, 40 fixed effect(s), and 6 variance component(s)(including residual variance). Calculating prior values of variance components by EM-REML ... Updated prior values: 0.000804336 0.0006864 0.000595213 0.000797577 0.0006809 0.000597679 logL: 10394.4 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. 1 11340.63 0.00047 0.00063 0.00054 0.00078 0.00056 0.00059 2 11898.92 0.00048 0.00061 0.00053 0.00073 0.00053 0.00056 3 11908.94 0.00048 0.00059 0.00052 0.00070 0.00051 0.00054 4 11913.10 0.00048 0.00059 0.00052 0.00068 0.00050 0.00053 5 11914.91 0.00049 0.00058 0.00052 0.00066 0.00050 0.00052 6 11915.72 0.00049 0.00058 0.00051 0.00065 0.00049 0.00052 7 11916.09 0.00049 0.00057 0.00051 0.00063 0.00048 0.00051 8 11916.41 0.00049 0.00057 0.00051 0.00063 0.00048 0.00051 9 11916.40 0.00049 0.00057 0.00051 0.00063 0.00048 0.00051 Log-likelihood ratio converged.
Calculating the logLikelihood for the model with the genetic correlation being fixed at 0.00000 Calculating prior values of variance components by EM-REML ... Updated prior values: 0.00042 0.00046 0.00049 0.00042 0.00045 logL: 10835.11030 Running AI-REML algorithm ... Iter. logL V(G)_tr1 V(G)_tr2 V(e)_tr1 V(e)_tr2 C(e)_tr12 1 10979.13 0.00000 0.00000 0.00063 0.00049 0.00056 (1 component(s) constrained) 2 -31260772.27 0.00000 0.00000 0.00063 0.00049 0.00056 (1 component(s) constrained) 3 -20690248.15 0.00000 0.00000 0.00064 0.00050 0.00057 (1 component(s) constrained) 4 -9656943.99 0.00000 0.00000 0.00064 0.00050 0.00057 (1 component(s) constrained) 5 -6939153.65 0.00000 0.00000 0.00068 0.00053 0.00060 (1 component(s) constrained)
Error: the information matrix is not invertible.
Analysis finished: Tue Nov 19 14:10:41 2013 Computational time: 0:2:28
|
|
|
Post by Jian Yang on Nov 20, 2013 2:53:34 GMT
An alternative test: (rg / se)^2 ~ chi-square(1)
I can see that you fitted a lot of covariates. What if you re-run the analysis without fitting the covariates.
|
|
|
Post by somatic on Mar 9, 2015 18:24:13 GMT
How can we estimate 95% confidence interval for rG? Do we need fisher transformation?
|
|
|
Post by Jian Yang on Mar 17, 2015 0:11:57 GMT
95% is approximately rG plus or minus 1.96*SE. I don't think Fisher transformation is necessary here.
|
|
abdel
New Member
Posts: 8
|
Post by abdel on May 11, 2015 15:39:17 GMT
--reml-bivar-lrt-rg 0 to test against rg = 0 --reml-bivar-lrt-rg 1 to test against rg = 1 --reml-bivar-lrt-rg -1 to test against rg = -1 What if there are 2 GRMs in the model? How do I specify weather I want to test rg1 or rg2? Thanks!
|
|
|
Post by Jian Yang on May 26, 2015 0:06:25 GMT
You are allowed to specify multiple values for the --reml-bivar-lrt-rg option, e.g. --reml-bivar-lrt-rg 0 0.5 is to test the null hypothesis that rg1 = 0 and rg2 = 0.5.
|
|