We aimed to compare mRNA expression of 2 genes in the human buccal epithelium quantified using the real-time PCR method in 2 time-point and in 2 groups. How I can calculate the minimum sample size? Whether is it enough per 10 patients per group (20 in general)? If yes, how ot can be proved.
As far as I understand, you need to calculate statistics in order to prove (or no) that there's a significant difference? In this case the bigger sample size you have - the better it is. However, statistical analysis might be performed starting from 4 samples in each group, so 20 specimens are enough. The main focus here is to use an appropriate statistical method.
I would argue that it is a nuanced argument, it is unclear to me whether an unpaired two-sample test would suffice here, we need an ANOVA, or something else? Number of samples is going to be dependent on the effect sizes and the noise that accompanies that signal. If you have large effect sizes, in this case large shifts in relative expression (i.e. 3-4 cycles), then you may not need so many samples to make a statement that treatment X causes large (3-4 cycles) shifts in expression. If you are looking for minor effect sizes i.e 1.2-fold change or something similar to that, then you would need a lot more samples than something that shows 3-4 cycles (or 9-16 fold change) of change (to obtain reasonable level of "confidence"). If you are working with a 1-factor scenario with something like a t-test (compare treated vs. untreated), G-power is a useful tool to explore the concept of statistical power. I would also encourage you consult with a statistician if you can, a well-thought out experimental design will save you a lot of heart (and head) ache later on! The test employed will also play a factor in determining the number of samples, a 1-way ANOVA with 4 groups is a little different to a two-tailed unpaired comparison.
https://www.psychologie.hhu.de/arbeitsgruppen/allgemeine-psychologie-und-arbeitspsychologie/gpower
The following R-package could provide useful if you decide on using any of the following tests: two-proportions z-test, a two-group comparison of Poisson data,comparison of negative binomial data, one-sample or two-sample t-test, t-test for non-zerocorrelation, Fisher’s exact test, signed rank test, sign test, rank sum test, ANOVA, and single-predictor Cox regression Article Computing Power and Sample Size for the False Discovery Rate...
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs