We want to calculate the sample size for a future study:
We have two groups called cases and controls
We want to set alpha value at 0.05
We want to set power at 0.8
We want a fold change of 1.5
What additional information do we need?
Any help or direction would be helpful.
Currently reading: http://www.nrcse.washington.edu/research/struts/chapter2.pdf
https://www.polyu.edu.hk/mm/effectsizefaqs/effect_size_equations2.html
You need to describe your experiment in more detail. Then look at this link. :
https://www.google.com/search?q=gene+expression+experiment+how+calculate+needed+sample+size&rlz=1C1CHBF_enUS847US847&oq=gene+expression+experiment+how+calculate+needed+sample+size&aqs=chrome..69i57.40082j0j4&sourceid=chrome&ie=UTF-8
At this point you should have the answer to your question. HTH David Booth
Hi there,
is the outcome continuous or categorical? Will you take into account baseline data to adjust for (as for an ANCOVA). What that 1.5 fold increase means? For example, if the test is an independent t-test (or even dependent t-test) it is important to know what the variability is (standard deviation). If you miss some elements you may use the standardized effect size https://en.wikipedia.org/wiki/Effect_size. Basically, given the statistical analysis you have to apply the same kind of approach to determine the power. Once all elements are clear it will be very simple for you to derive your power using G*power. You find manual and free download here http://www.psychologie.hhu.de/arbeitsgruppen/allgemeine-psychologie-und-arbeitspsychologie/gpower.html PS. 80% power is just acceptable may be better 90%, consider to increase the sample by possible lost to follow up... ;) )
STRUTS is a great resource. Already 2.1 gives you the answer. What you need is the Delta, which is the difference in means divided by the standard deviation.
But for gene expression this needs some extra information:
Gene expression is typically log-normal, so the log-concentration of the transcripts (or cDNAs) is approximately normal distributed. If you can show that the mean between the log-expressions is significant, it implies that the difference between (or the ratio = fold-change of) the mean expressions is significant.
So your data are log-concentrations (log expression values,possibly normalized, like dCt values from qrtPCR), and your task is a two-sample t-test on these values. You need to amke a guess what difference in these values you reasonably expect and what the standard deviation of such values will resonably be. This dependson the techology you use, the gene you are interested in and the experimental condition that is expected to change the gene expression.
To give an example, let'sconsider you want to do qrtPCR, so you measure dCt values. They are proportional to the log2 concentration, so a difference in dCt values of 1 means a fold-change of 2x. Further consider you study an infection and the target gene is a cytokine, which is believed to be strongly induced by infection, say you expect at least a 10-fold induction, what translates to a difference of log2(10) = 3.3 between the dCt values. The standard deviation of dCt values of independent samples is expected to be about 1.5 dCt. Now you can plug these numbers into the formula:
Delta = 3.3/1.5 = 2.2
n = 16 / Delta = 7.3
So you would need 8 samples per groupto get an 80% power at alpha= 5% assuming a 10-fold induction.
Thank you Jochen Wilhelm for your help. I must acknowledge the great answers, you've provided to gene expression research questions on ResearchGate.
another issue that is often important with gene expression data is correction for multiple comparisons. Typically in gene expression you have expression levels for many genes and so will be doing many comparisons to find the genes that best differentiate between cases and controls. If this is the case in your study then one approach is to use a sample size calculation that controls the False Discovery Rate. For this you need to specify how many genes you are testing and how many true positives you expect in the set. PASS has a procedure for doing this under 'microarray'. There are likely many other approaches to use. Probably best to consult a statistician or bioinformatician with experience in power calculations for gene expression or similar experiments.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size