[BioC] Fwd: covariate information

[Cholhee Jung]

 Dear users,

 Below is the question I posted originally in the 'ComBat user forum' of 
Google group.
 But, as I was suggested to forward my question to Bioconductor mailing 
list, I'm doing it now.

 Please find my question, below.


 Regards, 
 Chol-hee

On Tuesday, October 23, 2012 10:13:26 AM UTC+11, Cholhee Jung wrote:
>
>
>  
>  Dear users,
>
>  I was trying ComBat on ~1,000 samples.
>  Samples are spread over 12 batches and each batch contains 4 technical 
> replicates that are identical across all batches.
>  The number of covariates is 5, and I was using the ComBat implemented in 
> the 'sva' package.
>  
>  I tried ComBat with two model matrix built from the same covariate 
> information.
>
>  First model matrix was constructed as below:
>  >mod_mat = model.matrix(~as.factor(cov1) + as.factor(cov2) + 
> as.factor(cov3) + as.factor(cov4) + as.factor(cov5), data=pheno_data )
>
>  Second one was built as below:
>  >mod_mat = model.matrix(~as.factor(paste(pheno_data$cov1, 
> pheno_data$cov2, pheno_data$cov3, pheno_data$cov4, pheno_data$cov5, 
> sep=":")))
>
>  Basically, covariates were concatenated into one string for the the 
> second model matrix.
>
>  ComBat with the first model matrix raised the 'singular' error like below:
>
>   Error in solve.default(t(design) %*% design) : 
>    Lapack routine dgesv: system is exactly singular
>
>  But, ComBat run without error with the second model matrix.
>
>
>  Now I wonder if the two different model matrices are same?
>
>  Regards, 
>  Chol-hee
>
>