|
Post by michelatra on Jun 3, 2015 23:12:37 GMT
Hi Jiang I used bivariate reml model with 1 quantitative trait and 1 binary. I have 333 values for qt trait and 763 for case/controls (333 overlapping and 430 independent).
1)I used only overlapping individuals for both (333 values and case/controls) I used wo different commands once including --reml-bivar-nocove and once I calculated recidual covariance. The results are highly different and I need your help for the interpretation. The first with residual covariance included:
Source Variance SE V(G)_tr1 18.028821 11.676281 V(G)_tr2 0.046195 0.137366 C(G)_tr12 0.109697 0.910477 V(e)_tr1 4.729939 10.992880 V(e)_tr2 0.205741 0.133265 C(e)_tr12 0.318967 0.870802 Vp_tr1 22.758759 1.937684 Vp_tr2 0.251935 0.021028 V(G)/Vp_tr1 0.792171 0.488189: V(G)/Vp_tr2 0.183359 0.541333 rG 0.120203 0.959111 logL -593.716 logL0 -593.723 (when rG fixed at 0.000) LRT 0.014 df 1 Pval 0.4524 (one-tailed test) n 666
The second without residual covariance
Source Variance SE V(G)_tr1 19.000041 11.332581 V(G)_tr2 0.058078 0.133082 C(G)_tr12 0.440182 0.142308 V(e)_tr1 3.809812 10.634565 V(e)_tr2 0.194312 0.128821 Vp_tr1 22.809853 1.941344 Vp_tr2 0.252390 0.021075 V(G)/Vp_tr1 0.832975 0.470406 V(G)/Vp_tr2 0.230112 0.522271 rG 0.419033 0.501154 logL -593.782 logL0 -598.809 (when rG fixed at 0.000) LRT 10.054 df 1 Pval 0.0007601 (one-tailed test) n 666
Why do they look so different? Is the sample size too small?
2)I want to run a second model with qt trait for 333 indivs and a case/controls for 430 NOT overlapping indivs. Can I run a similar model or is better to include all 763 case/controls as trait2 and 333 qt as trait1?
Thanks
|
|
|
Post by Zhihong Zhu on Jun 5, 2015 8:06:09 GMT
Hi there, I think you could combine all the individuals together. Phenotypic value of binary trait for those samples from the quantitative trait is NA, similar to phenotypic value of quantitative trait for the rest samples. The big difference must be because of "--reml-bivar-nocove", which drops the covariance of residuals, as described in cnsgenomics.com/software/gcta/reml_bivar.html. Because the 333 samples are in common, you don't have to add this command. But when you combine all the samples together, i.e. phenotypes are from non-overlapping samples, you may need to do that. Cheers, Zhihong
|
|