Question: Need help setting up DESeq2 complex model for multi-dose experiment
gravatar for stwestreich
7 months ago by
stwestreich0 wrote:


I have RNAseq data from a longitudinal study, where six participants provided samples first at 0 dose, then at a low dose (25) of medication, and then at a high dose (35) of the same medication.  There are no massive changes that immediately jump out from the change in dosage, so I want to create a more complex DESeq2 model that will also look for significant changes occurring only in some of the individuals.

I've looked at the RNA time course vignette in the manual (, and while I want to look at that interaction term (interactions between dose and individual), I have more than 2 strains.  How can I take this into account?

If I only wanted to know about changes due to dosage level across ALL individuals, I'd run something simpler, like this:

ddsSimple <- DESeqDataSetFromMatrix(data_table_filtered, coldata, ~ Dose)


But I want to also capture (with a low p-value) any changes seen in one individual but not others.  Therefore, my current approach looks like:

ddsBMO <- DESeqDataSetFromMatrix(data_table_filtered, coldata, 
          design = ~ Individual + Individual:Dose + Dose)
dds <- DESeq(ddsBMO, test="LRT", reduced = ~ Individual )

1. Will this approach provide low p-values for individual-specific changes due to dose, namely where one individual shows a response to the increased dose even when other individuals reflect no change?

2.  If this approach isn't correct, how can I evaluate for individual-specific changes while still including all samples?

Many thanks for any help and guidance.

ADD COMMENTlink modified 7 months ago by Michael Love13k • written 7 months ago by stwestreich0
gravatar for Michael Love
7 months ago by
Michael Love13k
United States
Michael Love13k wrote:

I understood at first, 6 participants, each with three measurements. But then you mentioned strain, which I don't follow how that fits in. And you mentioned wanting to look at dose x individual interactions, which don't seem possible given that you can't produce biological replication from a single individual at a given dose. Can you show what the colData looks like so it is easier to understand the experimental design? 

ADD COMMENTlink written 7 months ago by Michael Love13k

Hi Michael,

Ah, my mistake - in the "fission" example in the linked vignette, the interaction term is looking for strain-specific interactions.  There's no "strain" term in my own data.

My colData looks like this:

Dose = rep(c(0, 0, 25, 35), times = 6)
Individual = rep(c("ind4003", "ind4005", "ind4006", "ind4008", "ind4009", "ind4016"), each = 4)
nameD <- colnames(data_table_filtered)
coldata <- data.frame(row.names = nameD, Dose = factor(Dose), Individual = Individual)

For biological replicates, I have, from each individual, 2 samples at 0 dosage (controls), 1 sample at dose level 25, and one dose level at 35.  Unfortunately, this is the only provided data, so I can't add more.

Given that I don't have multiple samples from each individual at each dose, does this mean that I can only measure changes across all individuals, not on a per-individual basis?

Thank you for your prompt reply and help.

ADD REPLYlink written 7 months ago by stwestreich0

Yes, I don't think you have enough replication to make individual-specific dosage inference, because you have no samples which capture the within-individual variation in expression after you start giving the medication. 

ADD REPLYlink written 7 months ago by Michael Love13k

Hi Michael,

That makes sense.  What if I treated the dose as binary, using a simple "0" or "1" for whether the individual received a dose?  In that case, I should be able to evaluate for individual-specific changes due to dose.

In that instance, would I use:

dds <- DESeq(ddsBMO, test="LRT", reduced = ~ Individual + Dose )

to look for individual-specific changes from consuming the dose?

ADD REPLYlink written 7 months ago by stwestreich0
It's ultimately up to you. It doesn't make sense to me to collapse two different doses into one, and have the difference in expression due to dose end up as unexplained variance.
ADD REPLYlink written 7 months ago by Michael Love13k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 325 users visited in the last hour