This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Design Formula With DESeq2

Hi,

I'm new to contrast / design functions with DESeq2. I have some RNAseq datas from MS patients. Here is an exemple of my sample table:

NAME    STATUS  CELLTYPE    COMPCHLES
S1.A    MS  CD4mem  LESLOW
S1.B    MS  CD4naiv LESLOW
S1.C    MS  CD8mem  LESLOW
S1.D    MS  CD8naiv LESLOW
S2.A    HV  CD4mem  HV
S2.B    HV  CD4naiv HV
S2.C    HV  CD8mem  HV
S2.D    HV  CD8naiv HV
S3.A    MS  CD4mem  LESHIGH
S3.B    MS  CD4naiv LESHIGH
S3.C    MS  CD8mem  LESHIGH
S3.D    MS  CD8naiv LESHIGH
S4.A    HV  CD4mem  HV
S4.B    HV  CD4naiv HV
S4.C    HV  CD8mem  HV
S4.D    HV  CD8naiv HV

I want to compare COMPCHLES (number of brain lesions) between my patients, depending on CELLTYPE and the disease STATUS. I tried something like this :

dds <- DESeqDataSetFromMatrix(countData = exprDat, colData = sampleAnnot, design = ~CELLTYPE + STATUS + COMPCHLES)

But its not working, and I dont realy get the following message:

Error in checkFullRank(modelMatrix) : the model matrix is not full rank, so the model cannot be fit as specified. One or more variables or interaction terms in the design formula are linear combinations of the others and must be removed.

If I can have some clues about this warning and some help to solve my problem, it would be nice.

Thanks, Hadrien

rna-seq
• 4,702 views • link
written by regue.hadrien • regue.hadrien

The rank of a model matrix, loosely speaking, is the maximum number of linearly independent columns to define the model. The info in your STATUS column is contained in the COMPCHLES column, as also told by swbarnes2 in the answer section. That is why, the model.matrix is not full rank and you need to drop at least one of them from the design formula.

• 0 views • link
written by Santosh Anand Santosh Anand

Thx for your anwsers and explainations, i'll check this quickly!

• 0 views • link
written by regue.hadrien • regue.hadrien

Well, Maybe I have an interpreting Question now.

In this case, using this formula :

dds <- DESeqDataSetFromMatrix(countData = exprDat,
                              colData = sampleAnnot,
                              design = ~CELLTYPE + COMPCHLES)
dds <- DESeq(dds)

And check results with:

res<-results(dds,contrast=c("COMPCHLES","LESHIGH","LESLOW"))

means check DE genes between "LESHIGH" and "LESLOW", not depending on the cellular type?

• 0 views • link
written by regue.hadrien • regue.hadrien

yes, that's the right interpretation

• 0 views • link
written by Santosh Anand Santosh Anand

Alright, thanks you.

Can I push a bit more? what should I type in design / contrast if I want to check DE depending on COMPCHLES, but in each cellular type? Maybe should I change my design?

• 0 views • link
written by regue.hadrien • regue.hadrien

since you are interested in the effect of the combination of CELLTYPE and COMPCHLES, you may create a new grouping as follows

newGroup = paste0(CELLTYPE, "_", COMPCHLES)

Then use only newGroup in your design formula. It's level will contain all the combinations of CELLTYPE & COMPCHLES. Then by using the contrast as before, you may pull the right contrast to compare.

• 0 views • link
written by Santosh Anand Santosh Anand

1 answer

Drop status from your design. The last column contains the same info, but more detailed.

That error means that you are asking the software to separate the effects of status and lesions, but that can't be done. See how all the HV status samples are also all the HV comp samples? The software sees that as hopelessly confounded. If this was a batch effect, there'd be no solution, but in your case, since the two columns mean much the same thing, you need to use only one.

• 0 views • link
written by swbarnes2 swbarnes2

Thanks, good to know, its working perfectly now!

• 0 views • link
written by regue.hadrien • regue.hadrien

Log in to answer this question.

More posts like this