Hypergeometric vs Binomial

Question

[yourA]

Hi,

I was wondering if there was any consensus on whether hypergeometric or binomial test is best suited to enrichment analysis of  a differentially expressed gene list. I know hypergeometric testing assumes dependency and is also suited to smaller sample sizes, where as the binomial is the opposite. Would I be right in saying that because  expression of genes can be dependent on other genes that this would better fit the hypergeometric test? Or would the fact I'm dealing with a large number of genes constitute using the binomial? What is the cut off for sample size? 

My guess is to use the hypergeometric test as most tools for this task utilize this, but I don't feel right basing my decision on this. Any help appreciated.

Answer 1

[Jean-Karim Heriche]

The binomial distribution corresponds to sampling with replacement. The hypergeometric distribution corresponds to sampling without replacement which makes the trials depend on each other. So you need to choose the one that fits your model. For differentially expressed genes, the correct model is the hypergeometric distribution. For a discussion of the tests to use in this situation, see this paper.