|
Post by completist on Sept 8, 2015 12:25:02 GMT
I have encountered some problems when using GCTA to estimate variances explained by all SNPs. I have about 900 samples and 800000 SNPs. Since the samples are getting from different sites, I used the site number as covariate. I used the data to estimate variances of 20 quantitative traits respectively and have some problems of the result. 1.I got the SE between 0.4 and 0.45. 2.For one trait h2 is nearly 0.000002 and for another trait h2 is nearly 1. 3.Some traits are related, however their heritability varied from a large range. 4.When I use 1 to 5 to code the different sites, it runs successfully, however when I use 5 covariates to code the sites with 0-1 variant, an error occurs.
Error: the X^t * V^-1 * X matrix is not invertible. Please check the covariate(s) and/or the environmental factor(s). I know that small sample size may cause some problems above. I want to know if there are other reasons causing problems above except small sample size. Thank you!
|
|
nickm
New Member
Posts: 2
|
Post by nickm on Sept 8, 2015 16:47:11 GMT
You only need 4 0-1 covariates to code for 5 outcomes, as the 5th outcome is a linear combination of the other four. Namely (1 - sum x_i).
|
|