When I fitted the null model in SAIGE step2 of single-variant test, it showed like following:
seqFitNullGLMM_SPA(pheno~age+sex, phenotye, grm, trait.type="quantitative", sample.col="IID", maf=0.005, missing.rate=0.01)
Filtering variants:
[==================================================] 100%, completed, 10s
Fit the null model: pheno ~ age + sex + var(GRM)
# of samples: 10,033
# of variants: 40,739
using 1 thread
Transform on the design matrix with QR decomposition:
new formula: y ~ x0 + x1 + x2 - 1
Start loading SNP genotypes:
[==================================================] 100%, completed, 13s
using 590.1M (sparse matrix)
Quantitative outcome: pheno
mean sd min max
0.2772192 0.05953586 0.05 1.04
Inverse normal transformation on residuals with standard deviation: 0.05748074
Initial fixed-effect coefficients:
x0 x1 x2
-5.064086e-09 8.959324e-05 0.0003868113
Initial variance component estimates, tau:
Sigma_E: 0.00165196, Sigma_G: 0.00165196
Iteration 1:
tau: (0.001737522, 0.001646519)
fixed coeff: (-5.064086e-09, 8.959324e-05, 0.0003868113)
tau: (0, 0)
Error in seqFitNullGLMM_SPA(pheno ~ age + sex, :
Sigma_E = 0, model not converged!
Does anyone have any idea about this issue? By the way, the phenotype and covariates data are clean without any outlier. Is it related to the SNPs for grm?
I just wish to solve this "not converged" problem. Thanks in advance!