I've run into various ways to average qPCR results.
Lets say I have populations A and B, both of which have several samples (from different individuals), all of which have been run with target and reference gene, with 3 technical replicates.
Assuming efficiency of 100%, to quantify expression difference in population A vs. B in target gene in comparison to reference gene I could
a) calculate expression (as 2^delta meanCq) for each sample, take arithmetic mean of these expression values (for each population & gene), then calculate ddCq for the population from those values
b) calculate expression and ddCq for each sample, then take arithmetic mean of these ddCq samples
c) calculate arithmetic mean of mean Cq's for the population, then calculate expression and ddCq. This option would actually equal option b if arithmetic mean is replaced by geometric one.
I have an uneasy feeling taking arithmetic means out of non-logarithmic variables, so I would go with option c.
What do you think?
I would probably take the mean of the technical Cq's, then with those, determine expression values for all genes (incl. reference gene) for the biological replicates. Then divide each GOI expression value by its own corresponding reference gene expression. Now you should end up with relative expression values for each GOI for each individual. Finally work out the mean of each population and the corresponding statistical significance.
I'm no expert in Biostat, but if I were you, I would just go for 2^ddCt of your target gene vs control gene in all your samples. Every individual is now represented by this value. Then, address the distribution of this value among your populations A and B. If you have a gaussian distribution, you may perform parametric unpaired t-test to adress the significance of the observed difference (the means are compared). If non gaussian, you may perform a non-parametric man-withney test to compare ranks (i.e. median) between your two populations.
Hope this helps!
Thanks for replies!
Gabriel: Yes, you're right. Actually I should have mentioned that I'm indeed using gene-specific (though not sample specific) gene amplification efficiencies. I just put that we're supposing perfect efficiency, because as far as I know, the issue I'm pondering (how to reasonably average results) is not really related to efficiencies.
Adam: Ok, good to know. So basically you're taking the (arithmetic) average of the relative expression values, the way I tried to outline as option b. As I said, I feel slightly uneasy about calculating the (arithmetic) mean of expression values, but perhaps that is indeed reasonable.
Antoine: Yes, that sounds reasonable, and indeed I think that the distribution of values lies in the heart of the problem. The problem is that (technical replicates aside), the number of samples in my population, the different fly-strains of a population, is typically low - say three, whereas replicates within the strain are easy to produce - which sounds kind of small for distribution or non-parametric test :). (Yes I know it's small for any test, but I would imagine some clear-cut differencies could be seen with parametric tests).
I'd like to use Cq differencies as such (or if efficiencies are to be taken into account, first compute expressions with known efficiencies and then take logarithms of those values), to account for exponential nature of gene expression, but so far with little support :)
- Badges
- Science method