Hi,

I have a dataset including 2 variables, pH (8 values) and strains consortium (5 values) all in triplicate. The pH variable can be seen as a time because I took samples during fermentation. I would like to answer two questions: 1- For a given strains consortium do I have gene expression differences at any pH point compared to my control pH point? 2- For a given pH point, do I have gene expression differences between my strains consortium compared to the control consortium? To answer these two questions I would like to know if it is possible with DESeq2 to block one of the 2 variables, for example comparing for the consortium of strain n°1 the differents pH points to the control pH or should I split my data set to keep only the results of counts of genes corresponding to the consortium of strain n°1 and use DESeq2 with a single variable, the pH?

I tried some models like :

dds <- DESeqDataSetFromTximport(txi, colData=colDesign, design= ~strain+pH)
dds <- DESeqDataSetFromTximport(txi, colData=colDesign, design= ~strain+pH+strain:pH)

But I can't get the desired comparisons : for pHcontrol :

controlstrain_vs_strain2          controlstrain_vs_strain3          controlstrain_vs_strain4          controlstrain_vs_strain5

for pH2 :

controlstrain_vs_strain2          controlstrain_vs_strain3          controlstrain_vs_strain4          controlstrain_vs_strain5

etc... for each pH value

And : for controlstrain :

pHcontrol_vs_pH2     pHcontrol_vs_pH3      pHcontrol_vs_pH4      pHcontrol_vs_pH5      pHcontrol_vs_pH6      pHcontrol_vs_pH7      pHcontrol_vs_pH8

for strain2 :

pHcontrol_vs_pH2     pHcontrol_vs_pH3       pHcontrol_vs_pH4      pHcontrol_vs_pH5      pHcontrol_vs_pH6      pHcontrol_vs_pH7      pHcontrol_vs_pH8

etc... for each strains consortium

I would like then generate one heatmap per pH comparison and per strain comparison since in each one I compare the samples to the same control. It would generate 13 heatmaps (8 with the pH and 5 with the strains consortium).

It is often easiest to combine two covariates into a single one, for example pH and strain becomes a column "ph.strain" and then you do ~ph.strain as design. Using contrasts (see vignette) you can then conveniently formulate any pairwise comparison that is required and extract results accordingly, e.g. for the heatmaps.