Error in Fit STAAR null model step: asMethod(object) : matrix is not symmetric [1,2]

[liu-xu-bing]

Hello,

When I ran the command in script STAARpipeline_Null_Model.r, I got this error: obj_nullmodel <- fit_nullmodel(phenotype~age+sex+PC1+PC2+PC3+PC4+PC5+PC6+PC7+PC8+PC9+PC10, data=phenotype,kins=sgrm,use_sparse=TRUE,kins_cutoff=0.022,id="sample.id", family=gaussian(link="identity"),verbose=TRUE) [1] "kins is a sparse matrix." Fixed-effect coefficients: (Intercept) age sex PC1 PC2 PC3 -0.09710790 0.01004128 0.01336379 0.59188157 0.82643259 -0.88178803 PC4 PC5 PC6 PC7 PC8 PC9 -0.38470585 0.74543827 0.04345983 0.81811801 -1.34958786 -0.52993129 PC10 0.13472704 Error in asMethod(object) : matrix is not symmetric [1,2]

Then I generated the sgrm data by FastSparseGRM with no error again and the result for command "isSymmetric(as.matrix(sgrm))" was "TRUE". I also check the colnames and rownames by running command "unique(colnames(sgrm) == rownames(sgrm))" and result was "TRUE". I don't know what lead to this error. Could you help me to resolve this problem?

Best Xubing

[xihaoli]

Hi Xubing,

Thanks for letting me know. Could you please check (1) whether your sparse GRM is symmetric, and (2) whether the rowname/colname of your sparse GRM has all sample ids for your phenotype/covariate file.

Best, Xihao

[liu-xu-bing]

I have tested both info and got the "True":

unique(phenotype$sample.id %in% colnames(sgrm)) [1] TRUE unique(colnames(sgrm) %in% phenotype$sample.id) [1] TRUE isSymmetric(sgrm) Error in UseMethod("isSymmetric") : no applicable method for 'isSymmetric' applied to an object of class "c('dsCMatrix', 'CsparseMatrix', 'dsparseMatrix', 'symmetricMatrix', 'dCsparseMatrix', 'dMatrix', 'sparseMatrix', 'compMatrix', 'Matrix', 'xMatrix', 'mMatrix', 'replValueSp')" isSymmetric(as.matrix(sgrm)) [1] TRUE

[xihaoli]

Hi Xubing,

Could you please then check (1) whether fit_nullmodel() works if you set kins <- NULL (assuming samples are unrelated); and (2) whether your sample.id in phenotype data and rowname/colname of your sparse GRM have duplicates (if there are no duplicates, then the dimension of phenotype and sgrm should be exactly the same).

Best, Xihao

[liu-xu-bing]

The result of first command was:

obj_nullmodel <- fit_nullmodel(phenotype~age+sex+PC1+PC2+PC3+PC4+PC5+PC6+PC7+PC8+PC9+PC10, data=phenotype,kins=NULL,use_sparse=TRUE,kins_cutoff=0.022,id="sample.id", family=gaussian(link="identity"),verbose=TRUE) [1] "kins is NULL, fit generalized linear model." It looks like code can run well without sgrm.

And the sgrm name was same to phenotype$sample.id, only different order between them:

length(colnames(sgrm)) [1] 641 length(phenotype$sample.id) [1] 641

[liu-xu-bing]

Hi Xihao,

I have tested another grm data which was generated by gcta tools and found that the script "STAARpipeline_Null_Model.r" runned successfully. I think maybe something error in old sgrm data which generated by FastSparseGRM.

In the end, thanks for your help.

Best Xubing

[xihaoli]

Hi Xubing,

Thank you very much for the updates. I can confirm that STAARpipeline supports any valid (sparse) kinship/GRM as input, so you can safely proceed to the next steps.

I shall close this case for now. Please let me know if you have any other questions.

Best, Xihao