Can you post the colData (the table where columns are variables and rows are samples). This makes it much easier to understand the design. You can do as.data.frame(colData(dds))

Here is the colData, although it is incorrectly formatted for DESeq2. As currently written, the controls (both treated and untreated with estrogen) are used in the "chemical" column. I believe what I need is for the controls to be represented instead as associated with a chemical at a concentration of 0.0. However, these would have to be duplicated due to the sharing of controls (if that makes sense). Here is a sample of the first ~15 rows (this isn't the best formatting):

         E2 Conc_uM rep chemical
PL1A_B02  0   1e+01   1      Tam
PL1A_B03  0   1e+00   1      Tam
PL1A_B04  0   1e-01   1      Tam
PL1A_B05  0   1e-03   1      Tam
PL1A_B06  0   0e+00   1  control
PL1A_B07  0   1e+01   1     PFOA
PL1A_B08  0   1e+00   1     PFOA
PL1A_B09  0   1e-01   1     PFOA
PL1A_B10  0   1e-03   1     PFOA
PL1A_C02  0   1e+01   1      Tam
PL1A_C03  0   1e+00   1      Tam
PL1A_C04  0   1e-01   1      Tam
PL1A_C05  0   1e-03   1      Tam
PL1A_C06  0   0e+00   1  control
PL1A_C07  0   1e+01   1     PFOA
PL1A_C08  0   1e+00   1     PFOA
PL1A_C09  0   1e-01   1     PFOA

As you can see, the controls are listed as a chemical. Let me know if I need to clarify further. Thank you for the help,

Rachel
 

Hi Michael,

Thanks for your response. To operationalize what you're recommending, would I create a new version of my colData table, where the rows labeled as "control" under the "chemical" column with a Conc_uM = 0.0 be changed to Tam? Will this limit my ability to analyze PFOA data because it won't know what set of results to use as a control? I'm thinking about when I do contrasts to compare PFOA at different concentrations to control or other scenarios as well.

I am going to want to do some dose-response analysis using these unevenly-spaced concentrations. Would you recommend that I create some sort of an ordinal index variable so that I can analyze how gene expression changes within a certain chemical across these concentrations. However, if I were to add that column to my colData table I'd run into the collinear problem again and get an error because two of the columns will essentially be identical. I currently have been using concentration as a factor

You mention "manually remove the interaction term of concentration=0 and chemical=PFOA (essentially saying that the concentration=0 level for Tam serves as the baseline for PFOA as well) from the model matrix before supplying this to DESeq()." What would this look like?

Thanks for your patience- I really want to gain an intuition for how DESeq is working so that I can use it to investigate all the contrasts we're interested in.

Rachel

Hi Michael,

Thanks again for all these answers. Here's what I'm confused about- doesn't each of the treated conditions need to refer back to a control condition to establish which genes are differentially expressed? In other words, if I turn each of the "control"-labeled conditions into "Tam" at concentration zero, how will DESeq2 know what to compare PFOA data to to determine differential expression? Furthermore, if I wanted to compare all the treated chemicals to control (to see whether any genes universally change in response to treatment when compared to control), how would this be accomplished?

I've looked through the vignette a number of times, but if there is a section that I should re-read to get a better sense of this, please do direct me.

Thank you,

Rachel

Hi Micheal,

Looking over this model matrix, why is replicates (rep) included as a covariate? 

Thanks,

-R