2. For molecular variation, this hierarchy of causation would look something like: drift + demography + genomic location + selection, with selection taking various flavors (purifying, divergent, balancing, etc)
Show this thread
1. Here's my take on the null model thing that includes ubiquitous forces like background selection and whether neutrality is an appropriate null. For (strict) hypothesis testing, it is good/necessary to have a nested series factors with obligate causation.https://twitter.com/pastramimachine/status/992447320624115712 …
-
-
-
3. As we know, just testing selection over drift (and mutation) can lead to a false results because things like rapid population growth (demography) can give a signature of positive selection in some tests if it is not properly accounted for.
Show this thread -
4. As
@pastramimachine and@3rdreviewer point out, the strongest global driver of molecular variation is genomic location as reflective of variation in recombination rate across the genome.Show this thread -
5. So under the model as specified, tests of selection should be carried out over the joint null of drift+demography+genomic location, not just over drift per site. In this sense, "neutrality" is not the appropriate null, although it for sure is a null for part of the model.
Show this thread -
6. If you assume that demography and genomic location (and other things) are strictly set to zero, which happens in the Neutral Theory, then you have drift/mutation as the null, whereas others might want background selection as the null via genomic location.
Show this thread -
7. To my mind "genomic location" has not be included in the general selection statistical testing framework in an appropriate global way yet, as it is thus far tends to be a very localized analysis still set again a neutral null background (e.g. Tajima's D).
Show this thread -
8. A more general approach would be to simultaneously include the effects of selection, linkage and recombination at all surrounding sites when analyzing any given site. However, we have a dimensionality problem that renders this practically impossible.
Show this thread -
9. So is background selection or something similar the "appropriate" null and neutral processes an inappropriate null? No, they are part of set of factors that can be sequentially built into a model, but it is true that test of selection should include genomic location in them.
Show this thread -
10. Now this is from a hypothesis testing, give me a damn p-value, point of view. Probably better to simultaneously fit the full model and estimate parameters and CIs so that we can start to think about relative contributions, which is what this debate is really about.
Show this thread -
11. It is important to not conflate "neutral processes" or a "neutral model" with the Neutral Theory, capital letters. The Neutral Theory can be false as a general explanation for molec var, but that doesn't mean that neutral processes aren't a necessary part of the null. FIN.
Show this thread
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.