Hi,
I have done 3 experiments (may be called biological replicates in this scenario) with the same cell line to measure gene expression of "XYZ" on days (D)4, 6, and 8.
For obtaining RNA, I have cultured cells in 6 wells (say replicates) for experiment-1. Pooled only 2 wells together at D4 and extracted RNA for qPCR analysis. qPCR experiment had three technical replicates. This way I got qPCR data for D4 for experiment-1. I did the same procedure for D6 and D8 to get qPCR data for D6 and D8 respectively for experiment-1.
In the same manner I repeated experiment-2 and experiment-3.
I wish to compare 3 time points D4 D6 D8 to see if increase or decrease of "gene A" expression is significant.
Is my data paired or unpaired? Could you please explain why so?
My assumption: My understanding is that the data is paired. Graphpad tests show some data sets to be distributed normally and some are not. Considering low sample number (n=3), I assume all the data to not follow gaussian distribution and therefore use non-parametric tests.
Can I compare with Friedman test with Dunn's post-hoc for XYZ at D4 D6 D8? Or should I use Wilcoxon testing two time points (like D4vsD6, D4vsD8 and D6vsD8)?
Hi Abhinav,
Considering the way you did the sampling, your data is not paired as you are taking different day's of samples from different wells. If you could take your samples from the same big dish at three time point, it would be considered as paired. Although some can argue that the original cells are from same passage but to be considered as 'paired sample' it has to be taken from same well(s) at multiple time points.
Now coming to 'Normal distribution' of your sample, this kind of samples you can always take as 'normally distributed' as the RNA is coming from at least more than 1000 cells. As per the 'Central Theorem limit' you can consider this as normally distributed samples. Don't just consider the samples as N=3, actually these are N=Couple of thousands/millions of cells. So, you can use parametric tests.
Hope, it will help you.
Thanks
Hi Jochen,I do not agree at all to this point that "The values on one plate are "paired", because they come from cells that were all seeded together and started growing together". That's not the right definition of paired samples. There are plenty of literature about it and similarly about the notions that why central limits of theorem applies to the RNA or Protein samples that are representing ei many of cells from a single well.
We can definitely argue on this matter citing many literature for both the statements.
Dear Jochen,
Please have a look at this. There are many of such-
A biologist's guide to statistical thinking and analysis*
David S. Fay and Ken Gerow.
For a quick look read "section 2.7. Is there a minimum acceptable sample size?"
Thank you Jochen and Debanjan for your insights.
As Debanjan said if samples can be taken from same big dish at different time points is valid but it could also be considered that the well plate itself is a big dish with partitions and so paired? There could be some variables like cells seeded in each well, how each well is handled, etc., but I took care to minimize them.
I'm not an expert in stats, Jochen. I'm not sure if gaussian and normal distribution are the same terms and I have said 'gaussian distribution' based on the GraphPad results. What I did is that I have entered the repeat dCt data in the GraphPad and have run the normality and log normality tests. The test result said whether a column has passed the test are not. If passed, I consider it to have normal distribution and therefore use parametric tests. Is this the right way to access data distribution?
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators