DESeq2 design comparing changes through 2 timepoints of 2 different treatments.

Question

[txema.heredia]

Hi,

I have RNAseq data of a experiment on cell lines as follows:

• 3 conditions: Control, TreatmentA, TreatmentB.
• 2 timepoints: 12h, 48h.

Initial assessment using pairwise comparisons between groups showed that both treatments up/down-regulate similar genes and pathways. However, the temporal dynamics between the 2 treatments are different.

I did also run a DESeq2 analysis using design = ~ timepoint * treatment . This allows me to obtain the comparisons timepoint48.treatmentA and timepoint48.treatmentB, both using control as the reference level.

I am interested in comparing the dynamics of the two treatments directly, using TreatmentA as the reference level.

However, I have doubts on the viability of such analysis: The samples are cancer cell lines kept in culture for 2 days. This causes even the control samples to show a ton of DEG between timepoints (794 down / 1002 up DEG). Thus, I want to study the temporal dynamics of the treaments, after removing the "changes through time" found in controls.

I am under the impression that, by running a design such as design = ~ timepoint * treatment, the timepoint part of the design is calculated on the reference level of the other variable. When using an interaction model with Control as the reference level for treatment, I get in timepoint_48_vs_12 802 down / 1006 up DEG. That is pretty close to the results using a simple pairwise model comparing Control48_vs_Control12 (794/1002).

---

Am I overthingking this? Is it possible to directly compare treatmentA vs treatmentB through time, but using only Control samples to correct for the "baseline effect" of time?

Should I limit my analysis to manually compare the results of timepoint48.treatmentA and timepoint48.treatmentB when setting Control as the reference level?

Or, as the temporal changes in Control samples ought to be shared between TreatmentA and TreatmentB, when using one of the treatments as the reference level, any difference between treatments (shared with control) will be interesting by either inclusion or omission and treatment-specific ?

Thanks in advance

Answer 1

[swbarnes2]

If you want to compare A to B, you should just compare them directly. If A increases 4 fold between timepoints, and B increases 3 fold between timepoints, that's a significant difference, whether or not it changes at all in the control.