Hi all,

I'm trying to normalize my Illumina MiSeq data using Phyloseq and DESeq2. I want to look at the interactions between my site location and by date. I attempted to use the following command:

Laurendds = phyloseq_to_deseq2(Lauren_data_merged, design = ~Location + Date + Location:Date)

But am getting this error:

converting counts to integer mode
Error in DESeqDataSet(se, design = design, ignoreRank) : 
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 

Doing some research into this issue I have seen that others have used the DESeqDataSetFromMatrix command (as in this post https://support.bioconductor.org/p/63134/),  but my metadata (mapping file) and my count table(.biom file) are already merged into my Lauren_data_merged object. I'm not sure where my variables are interacting to cause this error. I'll post an example of my metadata file below:

head(Thesis_mapping)
Sample Data: [6 samples by 11 sample variables]:
X.SampleID BarcodeSequence LinkerPrimerSequence InputFileName MonthSampleTaken Location Salinity WaterTemperature
PE12  NA NA PE12.fasta July M9 26.0 30
PE15 PE15 NA NA PE15.fasta July M9 14.5 30
PE17 PE17 NA NA PE17.fasta July BB 17.0 30
PE18 PE18 NA NA PE18.fasta July M9 17.0 30
PE20 PE20 NA NA PE20.fasta August BB 22.5 30
PE21 PE21 NA NA PE21.fasta August M9 22.5 30
Season Weather Date
 Wet Rain 7/10/13
 Wet Rain 7/19/13
 Wet Rain 7/23/13
 Wet Rain 7/23/13
 Wet Rain 8/3/13
 Wet Rain 8/3/13

hi,

Edited: It looks like you don't have replicates for fitting an interaction term here. To simplify, given the head of the colData which you have show, you have 2 locations and 4 dates:

loc date
 M9    1
 M9    2
 BB    3
 M9    3
 BB    4
 M9    4

You only observe once each combination of location and date, but you are trying to fit a model which has a different estimated mean for each combination, so there would be no way to estimate the dispersion.

You should remove the interaction term, and thereby control for base level location and date effects, or generate more replicates.

Do you have multiple samples for each combination of date and location? If not, then you can't fit an interaction term, because you will have as many coefficients as samples. You can reduce the degrees of freedom of the time variable using "ns" from the "splines" package. The limma User's Guide has a section on doing this: Section 9.6.2 "Many time points". I'm not familiar with how to use a custom design matrix in DESeq2, but you can certainly analyze your data in this way using either edgeR or limma-voom.