Questions About "Composite Of Multiple Signals"

Question

[zy041225]

Recently I read an article titled with "A Composite of Multiple Signals Distinguishes Causal Variants in Regions of Positive Selection". http://www.sciencemag.org/content/327/5967/883.abstract

I tried to implement the CMS but met some troubles.

1. what do the simulation work for? Was it used to distinguish the selected and non-selected region?

2. I saw lots of parameters in its supplementary, and I tried to use those parameters in cosi( a simulator ).However, I do not know how to add the additional parameter to get the different results. Besides, I don't understand how to judge whether the simulation works well or not. Which result of simulation is compared with real data, the neutral one or the selected one?

3. About the program they released at http://www.broadinstitute.org/scientific-community/science/programs/medical-and-population-genetics/cms/code. As I know little about python, I fail to run it, even the example pyTestCMS.py in the program.

Please help!

Answer 1

[Zev.Kronenberg]

The simulations are used to generate a NULL model of evolution. You compare the observed data to the NULL model testing for selection (p-values are derived here). The difficulty part, as you pointed out, is matching the population parameters ie (recombination, population mutation rate, generation time...etc...) to model your favorite population.

If you are working in human data there are many good resources for generating NULL model. Here are some of my favorites. The papers are worth reading.

MSMS: http://bioinformatics.oxfordjournals.org/content/26/16/2064.short

SelSim http://bioinformatics.oxfordjournals.org/content/20/18/3673.full.pdf

I am sorry to say there is not a painless way to learn the details of modeling genetic variation within a population. However, reading some primers in population genetics will be worth your time. You will be able to defend your model, parameters and conclusions by familiarizing yourself with the literature.

Good books:

Gene Genealogies, Variation and Evolution: A Primer in Coalescent Theory

Population Genetics: A Concise Guide

Good luck!

[zy041225]

In the article there are different kinds of simulation, based on one Neutral model, I think. The simulation generate 100 replicas per scenario, and then add one additional parameter to generate calibrated model. There are 90 different parameters, so there produce 90 different results. So I'm not sure how to judge whether the simulation is good or not. There's only one real data but 90 or 91 different simulation results.