Uncommon or Common Descent?


(Bill Cole) #150

If you look at figure 9 in Winston’s paper you see a module or family with at least 100 genes that is shared between rats and chimps but does not exist in humans and mice. This appears to be very unlikely from a perspective of random gene loss.

How would you explain this from and evolutionary perspective?


(Eric Michael Holloway) #151

If I’m following, Dr. Swamidass is claiming the DG results from gene loss, aka incomplete lineage sorting? Is this the only significant mechanism that is supposed to generate the DG?

I’m trying to get a grasp on what specific changes to the tree are supposedly resulting in a DG, since I believe they can easily be modeled mathematically, and we can get an intuition how helpful they are at explaining away Dr. Ewert’s result. Per my combinatoric analysis above, it seems like a whole lot has to happen to a tree before it resembles a DG.

And despite the 1.7% proportion, it is still the case the Bayesian factor states the DG is 2^{111,823} times better than a tree for explaining the data. That this is mostly an artifact of the dataset size is hard to swallow.


(Bill Cole) #152

Their can be deletions during meiosis that cannot be repaired. Having this type of deletion fixed in a population would seem to be a very rare event. I don’t think that unique genes are the result of ILS as that would show up as a variation of another gene if I understand ILS correctly.

Winston has identified a real problem with the tree in that it does not explain without rare statistical events gene families occurring outside but not inside taxonomic relationships.

Just thinking out loud I am not sure a tree would ever resemble a DG as they are very different ways to parse the data. The genes being mixed and matched outside the taxa is causing the DG to fit the data better.

[quote]And despite the 1.7% proportion, it is still the case the Bayesian factor states the DG is
2^111,823
times better than a tree for explaining the data. That this is mostly an artifact of the dataset size is hard to swallow.[/quote]

1.7% of the number of atoms in our universe is a boat load of atoms :slight_smile:


(S. Joshua Swamidass) #153

Before we go any further, let’s get this settled. We are not explaining away or dismissing @Winston_Ewert’s results. We are taking them seriously. We are treating him with respect, and are here to help.

If we cannot agree on this, I’m not sure explaining more is fruitful.


(Eric Michael Holloway) #154

However you wish to phrase it, you are minimizing the significance of the 2^{111823} number. If we take the number at face value, it means DG is categorically better than tree, no question about it. However, you claim the number can be produced due to something like p-hacking, where any difference between two theories becomes statistically significant with enough data, thus the number does not mean what it seems to mean. That’s why you throw out the 1.7% proportion as being more indicative of the result’s true significance.

You are also claim that common descent is dissimilar enough from a pure tree that the result does not directly imply common descent is incompatible with the data. However, the extremely large number makes this hard to believe, so you posit a variety of mechanisms that could cause common descent to diverge significantly from a tree structure.

Thus, there is a kind of explaining away of the immediate impact of the paper happening here. I believe you are doing this in good spirit and out of genuine desire to help, but at the same time you are countering the first impression someone comes away with after reading the paper.

So, my goal is to get some sort of “model of the model” that is clearer and easier to work with, which we can use to get an intuition whether the two minimizing factors you propose do in fact apply to the paper’s result. That’s why I proposed the combinatoric analysis, which at least shows the paper has identified the intrinsic weakness of tree like models from a graph theoretic standpoint.


(S. Joshua Swamidass) #155

Thank you for recognizing that.

Yes, that is what my job is, to offer the refined opinion of a person who knows this area extremely well. It is not surprising that my assessment will not match the first impression of someone new to this area.

So, I think you missed how the combinatorics argument hurts you. You are right, there are more directed graphs than trees. That means the penalty term is much greater for a directed graph. On a second read through his paper, it is not clear if @Winston_Ewert implemented this correctly. In fact, it looks like he made a subtle mathematical error. Now, I could be wrong, but it is silly to litigate such a subtle technical point on a blog. However, all these details have to be carefully worked out before anyone can have confidence in his findings.

Here is what we do know, that is a big deal for which to take credit. This is a good direction forward for ID. Winston is proposing an actual model to be tested with data. Even if it fails to bear out, it advances our knowledge. One reason to hold of the champagne is that if it does fail, @Winston_Ewert may have to retract parts of it. Such an event would not be a horrible thing, but is exactly what happens all the time in good science. Maybe this will work out. Maybe not. If not, maybe the next thing will.


(Eric Michael Holloway) #156

This seems the only route to show the result is not significant, that the math is wrong. None of the other possibilities I’ve seen mentioned sound like a plausible problem.

On the other hand, if the math is correct, then common descent is not a viable model for the data, since per my analysis common descent cannot be a significant enough divergence from a pure tree to impact the result.

So, we should focus on the math. If you have found an error, it’d be great to point it out.


(S. Joshua Swamidass) #157

Except this is very likely false. You have to prove it otherwise. Common Descent does not predict a perfect tree.


(S. Joshua Swamidass) #158

I’m not sure if it’s an error or not. The issue is subtle, and requires @Winston_Ewert’s input to clarify. It may be as simple that he did not clearly write out the methodology. We would need him to clarify. It might be an issue, or maybe not.


(Eric Michael Holloway) #159

What graph structure does Common Descent predict?

This is the problem, there does not seem to be a well defined prediction from Common Descent, so debating the point becomes fruitless.

As such, the dependency graph is the de facto preferred model since it is well defined, and fits the data better than the only other defined model, the tree.

So, the burden of proof is in the Common Descent court to define a specific graphical model and fit it to the data.

It is also unfair to expect @Winston_Ewert to do this, since Common Descent then becomes a moving target, a sort of gnostic knowledge only accessible to the high priesthood of evolutionary biologists.


(George) #160

Isn’t a question like this beyond the scope of the notion of Common Descent?

If you ask a dog breeder a question like this… he would orobably admit that a more spevific paradigm offers a better fit … but there is no notion that Common Descent is somehow invalidated! He knows everything about which dog is the parent or offspring … and that is the primary goal of Conmon Descent discussions.


(Eric Michael Holloway) #161

Then Common descent is not science.


(Eric Michael Holloway) #162

What is the supposed error?


(George) #163

@EricMH

I don’t really follow what you mean.

If I asked you what cancer has to do with theory of Common descent … and you said not much, does that mean cancer isn’t real or that dig breeders are faking the impression that they are raising a specific pedigree?


(Eric Michael Holloway) #164

Likewise. Science is a body of testable predictions. If Common Descent cannot make a testable prediction, then it isn’t science.


(George) #165

@EricMH

My point was not to say it isn’t testable.

My point was for the subject matter at hand , common descent is not really relevant.

For example… if we cannot predict rates of mutation… this doesn’t refute Common Descent.

Another example, the existence of convergent evolution ALSO does not refute Common Descent.


(Eric Michael Holloway) #166

I guess we disagree on that point. Common Descent is tree like, so it is fairly relevant. The question is where on the continuum between trees and dependency graphs does Common Descent lie? Too much on the tree side, then Dr. Ewert’s result refutes Common Descent. Too much on the dependency graph side, Dr. Ewert’s result does not tell us anything.

I argued previously a whole lot of edits have to happen to a tree before it resembles a dependency graph. This means many exotic behaviors have to be tacked onto stochastic variation + breeding to move Common Descent far enough towards a dependency graph to get a null result. Probably exponentially many as the dataset grows. While in itself this is already implausible, the multitude of addons also violates the principle of parsimony.

Of course, as you point out, if biological history is like dog breeding, and we already know without a shadow of a doubt the mechanism at work to produce the data, then the above is all a moot point. However, biological history is very much not like dog breeding, and thus we cannot assume off the bat that Common Descent is the best theory. We have to look at what model best fits the data we have, and a theory without a clear model is inferior to one with a clear model. Hence why the ball is in Common Descent’s court to make its case in light of Dr. Ewert’s research.


(George) #167

@EricMH,

Are you literally implying that even when there is no speciation involved, that Common Descent is to be questioned?

Can you give even one example that “Common Descent involving ONLY micro-evolution” should be questioned?


(Dr. Patrick Trischitta) #168

When discussing life origins, many theists will say that complexity implies a creator. They will ask how something so complex could happen by accident and use the diversity, variety and complexity of the natural world as a proof for intelligent design.

It seems that humans have pretty much always marveled at nature, at the complexities and diversity of life. The universe is spectacular; from the planets to bacteria, we seek to understand while simultaneously feeling as if we may never fully be able to. Our awe is often coupled with fear, fear of the unknown and fear of the realization that there is so little we can control. This awe and fear has led to the invention of creation myths - stories that attempt to explain how everything got here in the first place and how it continues to be sustained. We have developed many pathways toward understanding; we have learned, and continue to learn, so much.

Theists aren’t wrong when they say we still don’t have all the answers and that there is so much complexity that science still can’t explain. But where they are wrong, is in stating that this complexity necessitates a designer.

Creationists seem to be saying that an even more amazing being must be responsible for the amazing things around us. But if amazing, complex things need a creator, then who created the amazing, complex God? And if God is the exception, then theists are relying on certain “laws” (complex things need a more complex creator) to support their belief, but breaking the laws for those things that don’t fit.

It’s important to recognize that design and complexity are not the same thing. Looking at something complex and saying it requires design is begging the question and explaining a mystery with a mystery is really no explanation at all.

The theist will also often say, “The human eye is so complex, it clearly serves a distinct purpose; it was designed for something, and so there must be a designer. There’s no way that could happen by random chance.” And to a degree, they’re correct. Random one-time chance certainly did not form the human eye, but random chance and natural selection aren’t the same thing. Their argument betrays a broad misunderstanding of science, and evolution in particular.

Everything that has formed in the natural world, has done so as a result of natural processes following natural laws, repeated over billions of years. This is certainly not "chance," and while it is awe-inspiring, it in no way points to an intelligent designer.


(Ashwin S) #169

What you states is an assumption…
It’s truth is a matter of faith in your case…
With science, it’s a matter of methodology.
Both cannot dictate reality.