Question: Differential expression (DESeq2) miRNAseq on patient samples
0
2.6 years ago by
Bio_Ram0
Bio_Ram0 wrote:

Hello,

I have some miRNA raw counts of the patients samples before and after drug treatment. The problem is the blood samples were taken on different dates. Here are the fine details

Patient1: Before Treatment (BT), Day 14, 18,30

Patient 2: BT, Day 9,17

Patient 3: BT, Day 6,8

Patient 4: BT, Day 3,6

Patient 5: BT, Day 8,24

i tried using from the examples of DESeq2 for paired samples but unsuccessful probably because of different time points. Any help would highly appreciated.

modified 2.6 years ago by Michael Love25k • written 2.6 years ago by Bio_Ram0
Answer: Differential expression (DESeq2) miRNAseq on patient samples
1
2.6 years ago by
Michael Love25k
United States
Michael Love25k wrote:

The different sampling pattern potentially hurts your power, but there are still some options. Can you say what you might expect to see in miRNA expression as response to treatment? Eventually there would be saturation or return to baseline?

@Michael Love: I would definitely expect a differential expression in some potential miRNAs involved in the disease progression and more importantly their isoforms.

Yes, but does it increase over day and then stop, or go back down? These are important questions for modeling properly as you don't have a clean design.

It is really good question. From the raw counts, i could see few miRNAs which have really high count Time point 1 and drastically go down to "Zero" in the next time point and increases again in the next one. This kind of pattern were observed in most of the miRNAs. And also based on the onco gene expression at different time points, we presume even miRNAs have similar effect to control the expression of genes upon drug treatment.

It's going to be hard to come up with a flexible model without overfitting, because the grid is not regular, and you don't have so many patients. If I were you, I would define two time periods after BT, and discretize time into BT, T1 and T2. This way you can get a sense of the variability, but also allow that expression goes down or stays flat after T1 (similarly for down-regulated). You'll have to define where to draw the line between T1 and T2. You can either use some knowledge of the system or you can take a look at some general patterns across miRNA via exploratory data analysis.

It's important that patient (which you have as 1,2,3,4,5 above) is encoded as a factor and not as a numeric before running DESeq().

dds$patient <- factor(dds$patient)

Then you can make a variable time which is a factor with levels BT, T1 and T2 and use ~patient + time as the design.

After running DESeq() you can extract results tables:

results(dds, contrast=c("time","T1","BT"))
results(dds, contrast=c("time","T2","BT"))
results(dds, contrast=c("time","T2","T1"))