|
|
Post by Asling_GCTA on Dec 2, 2014 15:17:57 GMT
Hello,
I generated GRM based on genomic snps and I have a number of covariates as fixed effect in reml. However I also have another random effect, which is polygenic risk score needed in the model. I want to measure how much of the phenotypic variance can be explained by the score in the context of the existence of GRM. The matrix of this effect would have diagonal elements set to the value of the score, and off-diagonal elements set to 0. To estimate the proportion of phenotypic variance explained by the two random variance component, I created this matrix and fit this matrix with GRM in reml.
The result is:
V(G1)/Vp_L 1.927727 0.018416
V(G2)/Vp_L 0.008183 0.003945
G1 is the polygenic risk score component. I don't understand why the estimate of this random effect was larger than one. Maybe I didn't choose the right way to include random effect in GCTA. Can you please provide me some suggestions?
Thanks a lot!
Regards,
Asling
|
|
|
|
Post by guest on Dec 6, 2014 12:55:49 GMT
Hi Asling,
Are you saying you created another "GRM" matrix using the polygenic score?
1. I'm not sure whether the polygenic score can be fit as random effect, because S = sum(x*b_hat), S - polygenic score, x - coded genotypes, b_hat - estimated effect sizes
2. In GCTA, the regression model is y = x1b1 + x2b2 + ... + e, where X = {x1, x2, x3 ...} can be transformed to A matrix (GRM matrix). Please check your regression model.
|
|
|
|
Post by Asling_GCTA on Dec 9, 2014 10:40:48 GMT
Thanks for your reply! We actually want to test the effect of polygenic score on phenotype in the context of the existence of grm. We have tried to set the score as one of the covariate in model for reml. However we cannot find any option in GCTA to test fixed effect (something like wald test). That's why we are thinking that if we can treat polygenic score as random effect. Any suggestions about that? Thanks a lot! Asling Hi Asling, Are you saying you created another "GRM" matrix using the polygenic score? 1. I'm not sure whether the polygenic score can be fit as random effect, because S = sum(x*b_hat), S - polygenic score, x - coded genotypes, b_hat - estimated effect sizes 2. In GCTA, the regression model is y = x1b1 + x2b2 + ... + e, where X = {x1, x2, x3 ...} can be transformed to A matrix (GRM matrix). Please check your regression model. |
|
|
|
Post by Jian Yang on Dec 11, 2014 4:24:20 GMT
I think this can be done by fitting polygenic score as fixed covariate (--qcovar) in a GREML analysis.
|
|
|
|
Post by Asling_GCTA on Dec 11, 2014 9:39:40 GMT
I think this can be done by fitting polygenic score as fixed covariate (--qcovar) in a GREML analysis. We actually have tried to switch on --reml-est-fix option and put polygenic score as well as other covariates in --qcovar file option. In the output file of GREML we saw 2 columns: Fix_eff and SE. Do you mean that by calculating a t-score=Fix_eff/SE, we can test the significance of the estimate of fixed effect? The problem is that we don't know which row represents the test result for polygenic score. GCTA doesn't give the name of the tested covariate for each row. All we saw is two columns of numbers. Also we have covariates in --covar file, so we don't know if covariates in --covar were listed before or after the covariates in qcovar. Can I ask how does GCTA order the test results for covariates (both discrete and continuous variables) which are input in --covar and --qcovar? Thanks a lot! |
|
|
|
Post by Zhihong Zhu on Dec 11, 2014 11:54:50 GMT
Hi Asling,
1. The first row should be miu, population mean, 2. then quantitative covariates, 3. discrete covariates are at the bottom if you have.
Cheers,
Zhihong
|
|
|
|
Post by Asling_GCTA on Dec 11, 2014 16:51:18 GMT
Hi Asling, 1. The first row should be miu, population mean, 2. then quantitative covariates, 3. discrete covariates are at the bottom if you have. Cheers, Zhihong Hi Zhihong, Thanks! Can I take that the order in rows for quantitative covariates is the same as their input order? Asling |
|
|
|
Post by Zhihong Zhu on Dec 13, 2014 13:59:49 GMT
Hi Asling,
I'm sorry maybe I haven't correctly understand what you said. But yes, the order of effects of quantitative covariates in the .hsq file is the same as input order.
Cheers,
Zhihong
|
|
|
|
Post by Asling_GCTA on Dec 15, 2014 9:54:39 GMT
Hi Asling, I'm sorry maybe I haven't correctly understand what you said. But yes, the order of effects of quantitative covariates in the .hsq file is the same as input order. Cheers, Zhihong Thanks Zhihong! Now it is all clear for me! Regards, Asling |
|
