fitNULLGLMM: glmmkin.ai_PCG_Rcpp_Binary -> Get_Coef -> getCoefficients - spsolve(): memory allocation failure

[JonChan0]

Hi,

I am using SAIGE R package on a cluster node (>100 GB of RAM free) to try run step1: fitting the null GLMM using the following R script - saige_step1_submission.txt

However, this results in the following issue:

[1] "Genotype reading is done"
inital tau is  1 0.1
use sparse kinship to fit the model
Not enough memory to perform factorization.

warning: spsolve(): memory allocation failure
Error in getCoefficients(Y, X, W, tau, maxiter = maxiterPCG, tol = tolPCG) :
  spsolve(): solution not found
Calls: fitNULLGLMM ... glmmkin.ai_PCG_Rcpp_Binary -> Get_Coef -> getCoefficients
Timing stopped at: 130.1 1.001 131.5
Execution halted

This has been replicated across both UK Biobank's Research Access Platform using the Docker image: wzhou88/saige:1.2.0 as well as on UK Biobank's Research Access Platform's RStudio Workbench following install from source code, as well as on the aforementioned environment of local install (Conda) on another cluster node environment.

Full log attached below rescomp_saige_gene_errorlog.txt

[JonChan0]

Issue resolved with using different input files into step1.

[embertucci]

Hi @JonChan0 - would you mind describing what you changed in your input files to resolve this error? I'm running into the same issue. Thank you!

[JonChan0]

Hii @embertucci, I changed all input files:

• Sparse GRM with their Rscript with a different subset of individuals.
• Random markers used for variance ratio estimation using hard-called genotypes from those different subset.
• Phenotype file according to those different subset.

[embertucci]

@JonChan0 thank you for your quick response - much appreciated!