Stochastic Events and Natural Selection (Split thread)

[chadn737]

The problem is not looking, but finding, identifying, and in the context of describing Darwinian theory (as here) labeling - reliably.

 

This is true of any scientific experiment. Even if you know potential selective pressures, you still have a problem of reliably showing that they are indeed selective and that the selection on a particular allele is even caused by it in a first place. Error is inherent in all of science no matter how much information you have.

 

This is why we use statistics and set minimum thresholds that any test must pass before we call it significant.

 

About the stochastic events that happen to have altered various allele frequencies in populations - are you calling them "drift"?

 

 

By definition, that is what Genetic Drift is. Its the stochastic alteration of allele frequencies.

 

is not true of many of those events unless you have included them specifically and individually in your null model. How else are you differentiating them from selection pressure of an unknown kind?

 

 

You don't have to include them specifically. Neutral allele frequencies fluctuate stochastically, regardless of the source of that stochasticity. This can either be mathematically modeled or inferred directly from the sequence data itself. No assumptions about causative events are necessary.

[Arete]

is not true of many of those events unless you have included them specifically and individually in your null model.

 

You keep repeating this, however it contradicts population genetics quite comprehensively. As a result, you'll need to do more than simply state it.

 

You could start with something simple, like explaining why Tajima's D does not adequately account for stochastic variation. Here's the method:

http://www.genetics.org/content/123/3/585.full.pdf+html

and here's the paper where Tajima explicitly explores the impact of fluctuations in Ne on the statistic:

http://www.genetics.org/content/123/3/597.full.pdf+html

 

N.B. these papers are 25 years old, so there are more sophisticated methods out there, but proving inherent bias in Tajima's D would be a good starting point.

 

Does science claim to tell us about reality and the way things actually are? Or does science claim only to figure out what observations will obtain in certain situations? Does science explain observations or just interpret them using vocabularies that scientists have agreed to use for such interpretations?

 

The scientific method goes to great lengths to avoid absolute claims, however, when a theory allows for predictions to be made, and observations confirm those predictions, we are able to provide support for a theory. You don't just observe things and make up a story to explain them - you have to quantify how well they fit your predictions, and thus how well they support the theory behind them - and there's always probabilistic room left for the observations to fit the predictions through chance. Lengths are usually gone to through replication to minimize this chance, until we're pretty certain we have the best explanation for observed data.

 

 

Are we using "natural selection" with the claim that we're explaining a process that shapes allele distributions? Or do we just agree that "natural selection" is our shorthand term for allele distributions that conform to certain statistical parameters?

 

Again, the former. The invocation of natural selection allows us to make detailed predictions about the genetic makeup of populations. We then make observations of natural populations, and use statistical methods to quantify how well the observations match our predictions. In the case of quantifying selection in population genetics, the use of current methodology has been thoroughly validated with both theoretical and empirical approaches. In the words of xkcd - it works, bitches...

 

If you wish to make the claim that a prediction other than selection explains the observations more comprehensively, you'd need to validate that; you can't point at a p value of >0.001 and say "Hey look it might be wrong, so I reject it" objectively, unless you have a more probable explanation. In the case of selection and deviations from HWE, we have an extremely good fit of countless datasets, and no better explanation for the overwhelming majority of them. There are special cases where the assumptions of HWE are met and deviation is not caused by selection, but these cases are the rare exception, rather than the rule.

Edited December 3, 2014 by Arete

[overtone]

Neutral allele frequencies fluctuate stochastically, regardless of the source of that stochasticity

That's true by definition. The problem is determining whether or not some apparently selected allele was the beneficiary or victim of a stochastic, rather than selective, environmental event, thereby allowing one to label the allele "neutral".

You could start with something simple, like explaining why Tajima's D does not adequately account for stochastic variation

Why would I be concerned with that irrelevancy? He is addressing selection pressure on stochastic variations in the genetic pool of a normally reproducing population.

 

It doesn't even address the issue of detecting the influence of stochastic environmental events - random environmental influences - that happen by chance to have imitated environmental selection on specific alleles. It is no help in eliminating their influence from one's calculation of selection pressure, including them in one's null hypothesis, rewording one's description of Darwinian theory to account for their significant influence on the origin of species, etc.

 

 

 

and here's the paper where Tajima explicitly explores the impact of fluctuations in Ne on the statistic:

http://www.genetics....7.full.pdf html

If the problem were to estimate the general, expected, or average effects of frequent, significant, stochastic environmental events on a sufficiently large taxonomic grouping, that paper would come in handy.

Meanwhile, the subject of my posting was of course the influence of significant individual or small number stochastic environmental events on particular, narrowly described taxa - their propensity to imitate the effects of what you define as selection, in the very manner you have described as your detection and calculation criteria for selection pressure. So the paper is of little relevance to my posting, unless you can connect its insights somehow to what I am talking about - that seems possible, but as yet unattempted.

I'm assuming that somewhere in there we are all interested in the evolution of species, selection pressure as it influences the origin of species?

 

Specific to the topic of evolution:

Are we using "natural selection" with the claim that we're explaining a process that shapes allele distributions? Or do we just agree that "natural selection" is our shorthand term for allele distributions that conform to certain statistical parameters?

As you can see in Arete's links (146) they are in practice using the second, to guide or evaluate research. Darwinian theory of course rests on the first, as Ophiolite explained.

But the troublesome inconsistency there has no import for the theory itself (the particular events of selection that may have occurred in a particular case make no difference whatsoever to the basic theory, which merely posits the existence of some) and little to do with the validity of most results, explanations, etc, as published - because the actual researchers are in practice careful, argumentative, and generally humble in establishing specific mechanisms of selection or evolutionary histories. That is, you can't argue against Darwinian theory or its real accomplishments based on that essentially philosophical or terminological problem.

Or to put another way: they often work backwards, but establish their results forwards. Mathematicians, engineers, everybody does this. There's nothing wrong with it.

Edited December 3, 2014 by overtone

[Arete]

Why would I be concerned with that irrelevancy? He is addressing selection pressure on stochastic variations in the genetic pool of a normally reproducing population.

What? That's specifically what you're saying we cannot do; i.e. separate the genetic effects of selection vs stochastic change.

It is no help in eliminating their influence from one's calculation of selection pressure, including them in one's null hypothesis, rewording one's description of Darwinian theory to account for their significant influence on the origin of species, etc.

That's specifically what it does do....

 

If the problem were to estimate the general, expected, or average effects of frequent, significant, stochastic environmental events on a sufficiently large taxonomic grouping, that paper would come in handy.

Meanwhile, the subject of my posting was of course the influence of significant individual or small number stochastic environmental events on particular, narrowly described taxa - their propensity to imitate the effects of what you define as selection, in the very manner you have described as your detection and calculation criteria for selection pressure. So the paper is of little relevance to my posting, unless you can connect its insights somehow to what I am talking about - that seems possible, but as yet unattempted.

 

You have this backwards. Tajima's D statistic (and MK tests, Dn/Ds ratios, etc) apply to populations, not across taxonomic groups.

 

Edited December 3, 2014 by Arete

[chadn737]

That's true by definition. The problem is determining whether or not some apparently selected allele was the beneficiary or victim of a stochastic, rather than selective, environmental event, thereby allowing one to label the allele "neutral".

 

That is what all of these tests mentioned do. They determine if an allele is selective rather than stochastic. Alleles that fail to pass the threshold are assumed to be neutral rather than under selection. This happens to be the VAST majority of any genome. In humans, ~10-15% of alleles through the genome appear to be under selection, while the majority are not and thus neutral.

 

I would recommend that you take some time to learn about evolutionary and population genetics. Specifically the models of HWE and tests of selection. This debate really wont progress otherwise.

Edited December 3, 2014 by chadn737

[overtone]

Why would I be concerned with that irrelevancy? He is addressing selection pressure on stochastic variations in the genetic pool of a normally reproducing population.

What? That's specifically what you're saying we cannot do; i.e. separate the genetic effects of selection vs stochastic change.

I've clearly wasted every single minute I've spent posting on this matter here. WTF do I do now?

Soldiering on:

 

 

 

You have this backwards. Tajima's D statistic (and MK tests, Dn/Ds ratios, etc) apply to populations, not across taxonomic groups.

 

 

They apply to stochastic variation within a genetic pool while it is reproducing normally under what is presumed to be steady or sufficiently frequent given selection pressures, if any. On average, as they say. True if you have enough populations or large enough ones to average the outliers. And they are interesting, in that they tell us what to expect from the average population bottleneck, say, and it's not necessarily what people's intuition would have led them to expect - one of the pleasures of learning some probability and statistics.

 

So they are not relevant to any of my postings here. In the absence of some argument connecting them to my issues, they have nothing to do with them.

 

 

"That's true by definition. The problem is determining whether or not some apparently selected allele was the beneficiary or victim of a stochastic, rather than selective, environmental event, thereby allowing one to label the allele "neutral".

That is what all of these tests mentioned do. They determine if an allele is selective rather than stochastic. Alleles that fail to pass the threshold are assumed to be neutral rather than under selection. "

Which is the opposite of what I keep pointing out as the issue. You need to be able to screen for alleles that do pass the threshold, but due to the influence of events you don't want to include as "selective". Because in the world of evolutionary biology, that situation comes up fairly often.

Edited December 4, 2014 by overtone

[Arete]

So they are not relevant to any of my postings here. In the absence of some argument connecting them to my issues, they have nothing to do with them.

 

Back in post #115 you said this:

 

"That won't work, according to the professional understanding given us on this forum. You have to discover and subtract the influence of stochastic events - and no, a null model does not do that automatically."

 

So you told us that measuring changes in allele frequency and comparing them to a null distribution would not allow us to detect selection. You've never demonstrated why this is true, and now you're saying that a request to demonstrate that is untrue is irrelevant... leading this -

 

I've clearly wasted every single minute I've spent posting on this matter here.

 

To be about the only thing I would agree with that you've posted.

[chadn737]

Which is the opposite of what I keep pointing out as the issue. You need to be able to screen for alleles that do pass the threshold, but due to the influence of events you don't want to include as "selective". Because in the world of evolutionary biology, that situation comes up fairly often.

 

As we keep pointing out, these tests DO EXACTLY THAT. Tests like Tajima's D screen for alleles that pass the threshold.

 

In any experiment there is always a certain chance of false positives and false negatives...hence statistics. If you test an allele and it passes the threshold, on what basis do you reject it as being acted upon by selection? What you are suggesting smacks of cherry picking and confirmation bias....rejecting results that you don't agree with. The beauty of these tests is that they are not biased by any assumptions of what the selective force is or how adaptation should operate. If Evolution has taught us anything, its that the adaptations don't always fit how we expect them to. Sickle Cell anemia seems like a pretty bad adaptation, yet that is what it is. If you apply these tests to those alleles, they show that they have been selected for in those populations.

Edited December 4, 2014 by chadn737

[hypervalent_iodine]

!

Moderator Note

Update!* Thread(s) to remain closed.

* http://www.scienceforums.net/topic/85957-does-evolution-follow-the-scientific-method-if-so-how/?p=840232