elias
New Member
Posts: 9
|
Post by elias on Jun 6, 2014 21:59:23 GMT
Hi,
I used GCTA for over three years and I realized that the new version can't handle files in the same manner as the original version.
Here is an example:
I started running GCTA 1.24 on a quantitative traits with no covariates and I got this error message:
Error: the X^t * V^-1 * X matrix is not invertible. Please check the covariate(s) and/or the environmental factor(s).
However, when I run GCTA v1.04 - I got no error message but the output is clueless
Source Variance SE V(G) 955.077654 62.060516 V(e) 0.000486 45.725337 Vp 955.078141 33.530453 V(G)/Vp 0.999999 0.047876 logL -11831.269 logL0 -11830.108 LRT 0.000 df 1 Pval 0.5 n 3292
Now, I am wondering what does this error message mean:
Error: the X^t * V^-1 * X matrix is not invertible. Please check the covariate(s) and/or the environmental factor(s).
If I am not using any covariate but only one quantitative phenotype.
Similarly, this same phenotype work on a large sets of SNPs (11 million SNPs) but get this error message on a smaller sets of SNPs (700K SNPs):
Error: Log-likelihood not converged (stop after 100 iteractions). You can specify the option --reml-maxit to allow for more iterations.
But after modifying --reml to --reml-maxit 200 - the program do not even run...:
--mgrm XXX --pheno XXXX --reml-maxit 200 --out XXXXX --thread-num 3
Note: the program will be running on 3 threads.
Analysis finished: Fri Jun 6 14:09:40 2014 Computational time: 0:0:0
Thanks for your help
|
|
|
Post by Jian Yang on Jun 7, 2014 12:05:03 GMT
Re 1) It seems there is something wrong with the analysis. The LRT suggests that V(G) / Vp = 0 while the estimate (and SE) shows that there is a very strong genetic component. You can probably use the R script in GCTA website to read in the GRM in R to check whether there is something odd in the GRM (e.g. missing or abnormal values)
Re 2) I would check 1) and re-run 2). Of course, if the same size is small and you try to fit a model with many components, not all the parameters will be estimable.
|
|
elias
New Member
Posts: 9
|
Post by elias on Jun 9, 2014 21:58:58 GMT
Thanks for your prompt reply!
I still think there is a bug with the reml-maxit option When I use this option GCTA doesn't run at all
./gcta64 --mgrm-gz grm_chrs.txt --reml-no-constrain --pheno pheno1 --out Genotyped.SNPs --thread-num 3
******************************************************************* * Genome-wide Complex Trait Analysis (GCTA) * version 1.24.3 * (C) 2010-2013 Jian Yang, Hong Lee, Michael Goddard and Peter Visscher * The University of Queensland * MIT License ******************************************************************* Analysis started: Mon Jun 9 17:45:26 2014
Options: --mgrm-gz grm_chrs.txt --pheno pheno1 --reml-maxit 200 --out Genotyped.SNPs --thread-num 3
Analysis finished: Mon Jun 9
|
|
|
Post by Jian Yang on Jun 18, 2014 2:59:09 GMT
It's not a bug. You just need to add the --reml option in this analysis.
|
|