Entering edit mode
7 months ago
Jean Pierre
•
0
Hello everyone.
I am currently working with some RNA-seq data obtained from a previous work. The experiment consists of exposing a plant to a pathogen and then extracting the RNA at different times. I have the following samples:
- sample_1day - control_1day
- sample_7day - control_7day
- sample_14day - control_14day
In total I have 6 samples. As you can see there are no technical replicates because my objective is to compare the each time with respect to the each control. My question is the following, can I group the control samples to obtain 3 replicates and be able to perform analysis in DESeq ? I was thinking of doing some correlation analysis or something similar to indicate that the controls are similar. Recommendation on what would be your strategy to use in this case. Thank you for your time.
Your reasoning doesn't make sense - you'd need replicates to compare sample vs control for each time point. the only time you won't need replicates is if you don't care about statistical meaningfulness.
You will be able to ignore the time attribute and compare sample vs control using DESeq2 - the three time points will serve as replicates.
I understand, but in the research article where I obtained the data they make this comparison without replication. Therefore, I would like to know what could be done, would it be valid to join the 3 controls and compare with each time?
I understand that they would be inconclusive results
The missing replicates you refer to would not be technical replicates, but rather biological replicates. What you don't know, is the variation of a gene measurement in a plant on any given day. The only way to really estimate this is to have n plants, create and sequence n sample libraries, and then for any given gene you would have an estimate of the variation. Since you don't have that, and you seem to understand the caveat about uncertainty of any results, you can perhaps estimate the uncertainty for gene measurements using your control replicates. The edgeR vignette discusses estimating the uncertainty when you don't have replicates.
In your case, you have 3 "no treatment" replicates (control), meant in some sense, to provide the expression level of a gene in a plant under standard conditions on any given day (this is my assumption). What this doesn't control for is that lab conditions can vary day to day. Maybe your samples were collected in the field (which varies day to day). If your species is Arabidopsis, with a life span of weeks to months, your two week period across samples could account for 25% of it's lifespan (I would expect variation). If it's tobacco, with a lifespan of several months, I might expect less variation, depending where in the lifespan your samples were taken. It it's oak, with a lifespan of more than 100 years, I might expect samples taken a week apart to be a reasonable estimate of gene expression variation (still, the lab condition is not controlled for). And then keep in mind that none of this accounts for variation between individuals, when everything else is controlled for. Nonetheless, you can estimate the variation, which may help you explore the data and get some ideas, but will not be robust or defensible (and hopefully not a waste of time....though we never like our science to depend on hope).
What article is that? DESeq2 will not let you operate without replicates, so how did they manage to do this?