Your reviewer seemed to be picky, because the power is really not trivial to estimate
To estimate the power of a test, e.g. the t-test, you need to know:
- the type of the test (one/two sided, one/two sample,...)
- sample size
- the true difference in means between groups (for a really differentially expressed gene)
- the standard deviation (this is for one gene)
The true difference in means and standard deviation are normally not that easily available.
A simplistic (frequentist) way would be just to get the maximum likelihood estimate of the standard deviation from your data and average over it, and then setting the real difference in means to an arbitrary value (e.g. 2). Than you can claim with these parameters a t-test would yield this power if the data is normal.
In R you can use the function power.t.test for this simple calculation:
> example(power.t.test)
pwr.t.> power.t.test(n = 20, delta = 1)
Two-sample t test power calculation
n = 20
delta = 1
sd = 1
sig.level = 0.05
power = 0.8689528
alternative = two.sided
NOTE: n is number in *each* group
strict = FALSE)
That way you will probably vastly mis-estimate the variance and thus be overly optimistic about the power of your test, so the power.t.test function is more for illustration and understanding the concept.
More sophisticated methods I found:
The SPSA package in bioconductor.
You could also have a look at Black and Doerge (2002) and especially Page et al. (2006) who have implemented the PowerAtlas software for power analysis based on publicly available data.
Black MA, Doerge RW (2002) Calculation of the minimum number of replicate spots required for detection of significant gene expression fold change in microarray experiments. Bioinformatics 18(12):1609–1616
Page GP, Edwards JW, Gadbury GL et al (2006) The PowerAtlas: a power and sample size atlas for microarray experimental design and research. BMC Bioinformatics 7:84
+1 for doing power analysis BEFORE the study ... but I find that this is never the case.
The sizepower package looks more versatile than SPSA, as it works with many experiment designs whiel SPSA only works with two-sample comparisons.
I think Michael fixed my link - thanks :)
Depends which community I'm working with, the medics are used to doing it for studies involving patients, so expect to do it for their array studies as well (and referees in medical journals seem more inclined to ask for them if not stated). But it's not widespread amongst the biologists I work with.