|
|
Post by arineld on May 7, 2015 15:17:41 GMT
I am trying to run gcta on one binary trait (coded 0/1) and it gives me this error:
Error: can not find the 2th trait in the file [gcta.pheno.txt].
|
|
|
|
Post by Zhihong Zhu on May 10, 2015 3:17:57 GMT
Hi Arineld,
How many traits you specified?
The format of the phenotype file looks like,
FID IID pheno1 pheno2
1 1 1 1
2 2 1 1
...
100 100 0 0
When you assign "2" for the option "--mpheno", and only three columns are in the phenotype file, the program will be unable to find the trait. Please double check that. Thank you.
Cheers,
Zhihong
|
|
|
|
Post by arineld on May 26, 2015 10:08:05 GMT
Hi Zhihong and thanks for your reply.
I used the --mpheno option (--mpheno 1) and it still runs analysis for both trait 1 and 2. This is a bivariate analysis I am running on one trait. This is the output:
Source Variance SE
V(G)_tr1 0.029715 0.009501
V(G)_tr2 0.018095 0.004096
C(G)_tr12 0.001196 0.004422
V(e)_tr1 0.100667 0.003525
V(e)_tr2 0.038468 0.001444
C(e)_tr12 0.002681 0.001599
Vp_tr1 0.130383 0.006973
Vp_tr2 0.056563 0.003027
V(G)/Vp_tr1 0.227909 0.061345
V(G)/Vp_tr2 0.319913 0.056146
The estimate of variance explained on the observed scale is transformed to that on the underlying scale:
(Proportion of cases in the sample = 0.127851; User-specified disease prevalence = 0.045000 for disease 1 and = 0.045000 for disease 2)
V(G)/Vp_tr1_L 0.420148 0.113089
V(G)/Vp_tr2_L 1.477217 0.259259
rG 0.051579 0.190248
logL 9297.481
n 11201
|
|
|
|
Post by Zhihong Zhu on May 26, 2015 23:37:27 GMT
Hi Arineld,
The bivariate analysis is to estimate the genetic covariance or genetic correlation between two traits, which can be specified by "--reml-bivar", e.g. "--reml-bivar 1 2", estimating genetic correlation between the 1st and 2nd traits. Hence in the bivariate analysis, "--mpheno" doesn't work.
If you are going to estimate the SNP heritability, please use "--reml" instead, where "--mpheno" refers to the serial number of the trait, e.g. "1" - the first trait.
Cheers,
Zhihong
|
|
|
|
Post by arineld on May 28, 2015 10:34:45 GMT
Hi Zhihong and thanks again!
I am running the test on 1 binary trait (0 vs. 1), so estimating SNP heritability on one binary trait with one quantitative covariate and one discrete covariate.
I am using this command:
gcta64 --grm-bin /opt/file/grm/file_combinedrounds --reml --pheno gcta.binary_trait1.txt --covar gcta.03.covar.txt --qcovar gcta.02.qcovar.txt --out GCTA_trait1 --thread-num 6
|
|
|
|
Post by Zhihong Zhu on May 29, 2015 1:00:05 GMT
Hi Arineld,
Your command looks right. I may miss something, but do you mean that you still get error messages? Thank you.
Cheers,
Zhihong
|
|
|
|
Post by arineld on May 29, 2015 14:32:36 GMT
Hi again
Works fine now, thank you!
|
|
|
|
Post by arineld on Jun 2, 2015 9:50:06 GMT
Hi once again
This is the output I got from running this analysis:
Source Variance SE
V(G) 0.018094 0.004096
V(e) 0.038468 0.001444
Vp 0.056562 0.003027
V(G)/Vp 0.319894 0.056144
The estimate of variance explained on the observed scale is transformed to that on the underlying scale:
(Proportion of cases in the sample = 0.953303; User-specified disease prevalence = 0.300000)
V(G)/Vp_L 2.621375 0.460070
logL 5950.691
logL0 5938.908
LRT 23.566
df 1
Pval 6.035e-07
n 5632
This doesn´t look right, how can I interpret this?
|
|
|
|
Post by Zhihong Zhu on Jun 4, 2015 7:11:21 GMT
Hi Arineld,
I guess the disease prevalence would be 0.3% (i.e. 0.003) not 0.3.
Cheers,
Zhihong
|
|
|
|
Post by arineld on Jun 25, 2015 15:18:30 GMT
Hi Zhihong
Actually the prevalence of the "disease" or trait I was running is 30%...
However I am just wondering about what to do if I am running a binary gcta analysis and the prevalence is not known? In my sample the prevalence of the trait I am now checking out is around 50%. Do I just skip the --prevalence command?
|
|
|
|
Post by Zhihong Zhu on Jun 27, 2015 7:05:21 GMT
Hi Arineld,
Yes, you could skip that command.
Cheers,
Zhihong Zhu
|
|
