jcole
New Member
Posts: 4
|
Post by jcole on Sept 15, 2015 19:07:47 GMT
Hello, My dataset is of 3500 individuals for 48 variables in which I am predicting the total genetic effect using: gcta64 --remove excludeIDs.list --grm-bin test_Kibs --reml --pheno phen.phen --out LM1 --mpheno 1 --thread-num 10 --reml-pred-rand ... x 48 phenotypes The output is an HSQ for each variable which provides the genetic variance V(G), and the INDI.BLP which provides in column 4 the total genetic effect, which in theory is calculated from the genetic variance. However, when I combine all of the total genetic effects (column 4) into a single matrix 3500 x 48 and obtain the genetic covariance matrix using cov() in R, the diagonal (genetic variance) is not equal to the V(G) in the HSQ. The genetic variances from the HSQ files are in fact consistently higher than the genetic variance determined from the covariance matrix of the total genetic effect. Can you please explain this? Thank you!
|
|
|
Post by Zhihong Zhu on Sept 16, 2015 5:48:57 GMT
Hi there,
They are both estimates, and it is expected there would be slightly different.
Cheers, Zhihong Zhu
|
|
jcole
New Member
Posts: 4
|
Post by jcole on Sept 16, 2015 16:27:40 GMT
I've edited my question with a graph, hoping to show that there is a consistent pattern of decreased genetic variance in the covariance matrix of total genetic effect.
Can you explain this a bit further for me? Why would this method have lower estimates in particular? Are there some differences in assumptions in the methods and what are they?
Thank you for any clarification you can provide!
|
|
|
Post by Jian Yang on Sept 17, 2015 0:41:52 GMT
The reported values in INDI.BLP are the estimated (or someone will call it "predicted") genetic values. So, they are g_hat. The reported value s in the .HSQ file are the genetic variance, i.e. var(g). var(g) != var(g_hat) as shown in your plot.
|
|