[BioC] LIMMA package - coefficient not estimable

[Sean Davis]

On Fri, Mar 30, 2012 at 8:09 AM, yeeling chong <yeelingchong at gmail.com> wrote:
> Dear Lists,
>
> I'm currently using LIMMA package to analyze my microarray datasets.
> For each treatment, I have three biological replicates. The subset of
> the constructed design look like this:
> Array
> R16A    R16B    R16C    R16D    R16E    S03A      ------   Biorep1      Biorep2 Biorep3
> 02152012_462157_A01_1613_cy5_R16E       0       0       0       0       1       0       ------ 1        0       0
> 02152012_462157_A01_161_cy3_R16A        1       0       0       0       0       0       ------ 1        0       0
> 02152012_462157_A02_81_cy5_R08A         0       0       0       0       0       0       ------ 0        0
> 02152012_462157_A02_813_cy3_R08E        0       0       0       0       0       0       ------ 1        0       0
> 02152012_462157_A03_164_cy5_R16B        0       1       0       0       0       0       ------ 1        0       0
> 02152012_462157_A03_1614_cy3_R16E       0       0       0       0       1       0       ------ 0        1       0
> 02152012_462157_A04_89_cy5_R08C 0       0       0       0       0       0       0       ------ 0        1
> 02152012_462157_A04_811_cy3_R08D        0       0       0       0       0       0       ------ 0        1       0
> 02152012_462157_A05_1611_cy5_R16D       0       0       0       1       0       0       ------ 0        1       0
>
> When I fit my data into lmFit, it doesn't estimate the coefficient of
> the third biological replicates. The error msg looks like this:
>
>> Fit=lmFit(norm.dataM,designlm,correlation=corfit$consensus)
> Coefficients not estimable: Biorep3
> Warning message:
> Partial NA coefficients for 22664 probe(s)

The coefficients not estimable means that you do not have enough
degrees of freedom given your number of samples and number of effects
you are trying to estimate.  In the normal use case, one does not
include the BiorepX columns in the design matrix.  I'd suggest
removing the Bioreps columns from the design matrix (unless you really
are interested in the biological replicates).

Sean



> Can anyone tell me what's goes wrong my my design? Thanks a lot
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor