After tesets of selection on WGS data, there would be a lot of regions as significant in at least one test under the correspongding 0.1% genome-wide FDR. And my question is:
How to prioritize regions for the subsequent research?
There are many frameworks to combine the results of multiple tests for positive selection, identifying only those highlighted by more than one method. However this is a complex problem, complicated by the fact that each selection test measures a slightly different type of selection event (e.g. complete or incomplete sweeps, recent or ancient, population-specific sweeps, etc..)
The easiest way is to combine the p-values using Fisher's method. This is not completely correct because the p-values from multiple tests are not always independent, but in general it is accepted. For a discussion, see:
- Peng et al 2010 http://www.nature.com/ejhg/journal/v18/n1/full/ejhg2009115a.html
- Zaykin et al 2008 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2569904/
- This biostar discussion How To Combine P-Values Obtained With Different Approaches?
- wikipedia article http://en.wikipedia.org/wiki/Fisher%27s_method
A more sophisticated approach is to use machine learning to combine the tests together. However, this usually requires calculating simulations using different demographic models, in order to have a training and test datasets.
- Grossman 2010, CMS http://www.ncbi.nlm.nih.gov/pubmed/20056855
- Lin et al 2011, Boosting http://www.ncbi.nlm.nih.gov/pubmed/21041556
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
It all depends on what kind of research you want to do afterwards...
The mission is to find out genes under selection. But I found there were too many genes significant in at least one test...What do you recommend, such as some experience or papers?