DESeq2 Circadian Time Series without time 0

[jaw_bio]

Hello all,

I am using DESeq2 for the following analysis:

I have RNA-Seq data from 2 tissues and 4 time points, with 4 replicates for each condition. I have two tissues (liver and scn) and four timepoints (3,9,15,21).

I'm trying to test for genes where time is a factor, so my full model is

~ tissue + time + tissue:time

while my reduced is simply

~ tissue

Things look good, but when the results get returned, the model has compared each timepoint to time 3:

> resultsNames(dds)
[1] "Intercept"           "tissue_scn_vs_liver" "time_9_vs_3"         "time_15_vs_3"        "time_21_vs_3"       
[6] "tissuescn.time9"     "tissuescn.time15"    "tissuescn.time21"

Since I'm looking for evidence of circadian regulation, time 3 isn't really a baseline starting point as in the fission data from the Bioconductor rnaseqGene workflow.

I'm thinking pairwise comparisons across time points might be better, even though the number of points is small (4).

I'd appreciate any help in building my model better.

Thanks!

-John

[dpryan]

I would suggest using an LRT with a reduced model of "~tissue+tissue:time". That would directly answer the question, "After accounting for tissue-specific differences and interactions, what genes vary with time?" It seems that that's what you're asking. This should expose the typical genes, such as core clock members used in both the SCN and liver.

[jaw_bio]

I would suggest using an LRT with a reduced model of "~tissue+tissue:time". That would directly answer the question, "After accounting for tissue-specific differences and interactions, what genes vary with time?" It seems that that's what you're asking. This should expose the typical genes, such as core clock members used in both the SCN and liver.

Thanks! I appreciate the response.