I am trying to fit a linear mixed effect model with correlated phenotypes (pre-transformed to residuals already). But I am receiving the error message saying "GSL ERROR: matrix is singular in lu.c at line 449 errno 1".
I checked the previous posts in this channel and found someone mentioned the issue may be triggered by the correlation in the phenotype/covariates. But it does not make sense to me given the related publication of this function mentioned the use of lmm on correlated phenotypes. I also tried with the example data posted on Github with the fully duplicate pheno1 and pheno2, and it works.
Thus, I am not sure why this issue comes out in my analysis. Can anyone help me out? The full log file is pasted below. I really appreciate.
----------------------------------------------------------------------------------------------------------------------------------------------./gemma -bfile ./Data/Genotype/ABCDS_multitrait_Set1_AB -k ./Data/Genotype/ABCDS_multitrait_Set1_AB.sXX.txt -lmm 2 -p ./Data/Phenotype/ABCDS_ab_pheno_residuals_no_header.txt -n 5 6 -o ABCDS_multitrait_Set1_AB_output.txt
GEMMA 0.98.5 (2021-08-25) by Xiang Zhou, Pjotr Prins and team (C) 2012-2021
Reading Files ...
## number of total individuals = 199
## number of analyzed individuals = 199
## number of covariates = 1
## number of phenotypes = 2
## number of total SNPs/var = 8795124
## number of analyzed SNPs = 8576519
Start Eigen-Decomposition...
REMLE estimate for Vg in the null model:
7.1255
0.0694 0.0007
se(Vg):
0.7822
0.1872 0.3842
REMLE estimate for Ve in the null model:
0.0009
0.0262 0.8702
se(Ve):
0.0026
0.0649 0.3752
REMLE likelihood = -739.4533
MLE estimate for Vg in the null model:
7.1668
0.7299 0.1201
se(Vg):
0.7216
0.1905 0.0879
MLE estimate for Ve in the null model:
0.0007
0.0232 0.7598
se(Ve):
0.0000
0.0000 0.0000
MLE likelihood = -716.3921
GSL ERROR: matrix is singular in lu.c at line 449 errno 1
