Mud-to-Mozart Atheology (Or, Who are the real skeptics?)

_{Gil Dodgen

October 22, 2011

Intelligent Design}

Share: Facebook; Twitter; LinkedIn; Flipboard; Print; Email

I find the “skeptic” claim on the part of Darwinian materialists very interesting and equally illuminating. Darwinists exhibit no skepticism whatsoever about the thesis that physical stuff turned into Mozart by chance. (Don’t try to deny this, Darwinists, that is the essence of your claim. You can try to obfuscate with legion “peer-reviewed scientific papers,” but you’re not going to fool me and many others about what you are actually promoting and advocating.)

I choose Mozart not just because I am a classical concert pianist, but because his existence epitomizes everything that Darwinian theory is totally powerless to explain.

Darwinists, claiming to be skeptics, actually exhibit the antithesis of skepticism — making transparently ludicrous claims and providing a never-ending stream of unsupported extrapolations, based only on wildly imaginative speculation with no empirical support.

How is it that Darwinian atheists are the only ones who get to declare themselves legitimate skeptics? Is mud-to-Mozart-by-chance philosophy the only worldview immune to skeptical inquiry?

Comments

Natural selection does not act nor select. Not only that with AVIDA it has been demonstrated that once realiistic parameters are set your position looks like crap: The effects of low-impact mutations in digital organisms Chase W. Nelson and John C. Sanford Theoretical Biology and Medical Modelling, 2011, 8:9 | doi:10.1186/1742-4682-8-9
Abstract: Background: Avida is a computer program that performs evolution experiments with digital organisms. Previous work has used the program to study the evolutionary origin of complex features, namely logic operations, but has consistently used extremely large mutational fitness effects. The present study uses Avida to better understand the role of low-impact mutations in evolution. Results: When mutational fitness effects were approximately 0.075 or less, no new logic operations evolved, and those that had previously evolved were lost. When fitness effects were approximately 0.2, only half of the operations evolved, reflecting a threshold for selection breakdown. In contrast, when Avida's default fitness effects were used, all operations routinely evolved to high frequencies and fitness increased by an average of 20 million in only 10,000 generations. Conclusions: Avidian organisms evolve new logic operations only when mutations producing them are assigned high-impact fitness effects. Furthermore, purifying selection cannot protect operations with low-impact benefits from mutational deterioration. These results suggest that selection breaks down for low-impact mutations below a certain fitness effect, the selection threshold. Experiments using biologically relevant parameter settings show the tendency for increasing genetic load to lead to loss of biological functionality. An understanding of such genetic deterioration is relevant to human disease, and may be applicable to the control of pathogens by use of lethal mutagenesis.
And if you have a duplication even then all you have is another protein diffusing through the organism ready to get in the way of important reactions.Joseph_{November 1, 2011
November
11
Nov
1
01
2011
04:24 AM
4
04
24
AM
PDT}

a) NS, both negative and positive, seems to have no role, or only a really minor role b) Neutral variation can easily happen, and can easily reach any possible target
Of course NS has a role - thats what EL and I keep saying - neutrality is DEFINED by NS acting on the search space. NS acts to keep a sequence neutral when there is no hill to climb but flat areas exist. Neutral variation cannot easily reach any target - how could a sequence of neutral mutations reach a target in the middle of a sea of deleterious function? They can't because that, as a statement, makes no sense. By definition a move into deleterious function is not neutral!
This is why you have to consider both the search space topology
Sure. And the search space topology is extremely simple. The greater the different between two sequences, the greater the distance. Very simple, isn’t it? There is not much to be considered.
Sorry but that is nonsense - Topology refers to the shapes of the surfaces in the landscape. What you are talking about is just the distances between two points. Consider these three examples: 1 If there exists a series of steps from A to B, and each confers an advantage, then an evolutionary 'search' can 'climb the hill' 2 If there exists a series of steps from A to B, and each confers no advantage or disadvantage, then an evolutionary 'search' can 'cross the ridge' 3 If there exists a series of steps from A to B, and each is seriously harmful, then an evolutionary 'search' will never take more than one step in that direction. In the first two examples above the topology provides a navigable route from A to B, but not in example 3. Therefore the probability of getting from A to B in the three examples is different in each case (1 is the highest, 3 is the lowest) despite the distance between A and B being the same.
Free neutral search, like in duplicated genes or non coding DNA, can search the whole space. It just needs time to get away from the initial state (not too much, indeed).
No. You are still making the same cognitive error as before in viewing neutrality as different than selection. The sequence space includes sequences that code for proteins, that are not expressed, that are non coding and that are deleterious. Only those areas of the sequence space where a point mutation will not cause a change in selective potential are neutral, the rest are not - the rest are areas of sequence space upon which NS will act.
The distribution of probabilities for achieving any particular state from a given point and in a given time are highly dependant on how your search proceeds and the shape of the landscape you are traversing.
Sure. And that’s exactly what I have tried to analyze in detail.
Ok, that is confusing because you said this above:
The whole dFSCI barrier applies
and previously you have been saying this:
The problem is, if the system is only random, or mainly random, the probabilities are those relative to the whole dFSCI of each protein.
and this:
But now it is perfectly true that the search space can be freely traversed by NS, because there is no more the limit of negative selection, defending the functional island of A. It is true, as you say, that the initial variation will remain local, but as neutral variation accumulates, we are diving into the sea of unrelated states. And we can, now, reach B. But the probabilities are exactly those deriving from the total dFSCI of B. Or of the functional internediate, if and when it exists.
Both of those statements are about the probabilities of getting from A to B using random sampling (or at best a random walk), not evolutionary search. They seem to assume that the entire search space is neutral with respect to fitness. The entire sequence space encompasses neutral, deleterious and advantageous areas, some of these areas do not form part of the searchable space - If you want to analyse the probabilities of getting from A to B in an evolutionary system you need to have a good understanding of the topology of the fitness landscape, you need to take into account how much of the sequence space is not searchable, how much of it is neutral, and how much of it confers some advantage. Just computing the distance from A to B tells you nothing in this regard. I think the error you seem to be making is in thinking that neutrality refers to a different search space than non neutral - it doesn't, it is just a feature of the fitness landscape for the entire search space of the sequence. A neutral mutation does not render the entire search space neutral. If you have, for example, a duplication event, then a mutation to the duplicate will not destroy the function that exists in the original. The search space for the copy includes the existing function (a protein domain), other new functions (other protein domains), neutral and advantageous non coding sequences, non expressed sequences and deleterious sequences. The question here is whether there exists a route through sequence space for the duplicate from its starting state to a new functional state that does not require a move into deleterious function. Neutral moves don't, and neither do moves into new advantageous functions of either a protein or non protein coding sequence.DrBot_{November 1, 2011
November
11
Nov
1
01
2011
04:02 AM
4
04
02
AM
PDT}

EL:
Sure, it’s a metaphor, but it’s a very precise metaphor, and can be translated into real productive ouputs, as DrBot will attest.
In what ways is it productive? EL:
Meaning what? Yet again, it isn’t clear when you are talking about phenotype and when you are talking about gene! And if you are talking about a gene, then it matters whether you are talking about sexual or clonal reproduction. And also whether you are considering HGT.
Why add needless complications? The main distinction is between "highly conserved" regions of the protein's gene sequence, and those parts that are not. EL:
Yes indeed, by gene-gene interaction, but I’m not sure that’s what you mean. Perhaps you mean that it’s possible for a previously neutral phenotypic feature to become beneficial. Which is also true. But it’s important to distinguish between the two, as they are quite different.
Why not simply talk about genotypes? Whether the fitness involves gene-to-gene interactions or whether it involves strictly phenotypic features doesn't really matter than much when one is trying to talk about general principles. Yes, each might have it's own special considerations, but the basic principles underlying the one or the other remains much the same. Again, why the needless complications? EL:
Well, of course, but that’s looking from the evolved protein backwards, not from no-protein forwards. Once something has become useful, then you are right, there is less “wiggle room” – which is why gene duplication is probably an important process, and also HGT, in generating novel selectable phenotypic variance.
Is this the spandrels of St. Mark's here? Are we talking "exaptation"? If so, this is an example of a "just-so" type scenario. The problem with Darwinism is the lack of intermediates in the fossil record. The problem with neo-Darwinism is the lack of molecular intermediates: we don't see them, but we'll just assume they happened nevertheless. You quibble with Behe and Snoke's paper. Well, here's part of what they say in response to Michael Lynch's criticism: "This is a strong requirement—that not only the end products, but steps along the way to a multiresidue function, must be either selected or at least neutral. Michael Lynch makes a similar assumption. Our model posited necessary intermediate mutations to be deleterious in the unduplicated gene; Lynch’s model assumes them to be neutral: ‘‘all 20 amino acids are equally substitutable in the intermediate neutral state’’ (Lynch 2005, this issue). All of his objections to our work stem from this difference." This assumption on Lynch's part seems to directly contradict what you just stated: Once something has become useful, then you are right, there is less “wiggle room”. Lynch allows all the wiggle room imaginable in the unduplicated gene. EL:
But being highly conserved after-the-fact is not the same as being vital before there is such a thing as the protein in the first place. Again, this is one of the many fallacies in Behe’s concept of Irreducible Complexity.
Here we go again with the miracle of "exaptation"! Has "exaptation" been "demonstrated", or is it simply "assumed" to have happened? Where's the proof? Or is it another "just-so" story once again? EL:
If you are talking about DrBot’s gene-space, the values are phenotypic fitness, and if there is a big “cliff” then that’s not a problem – the phenotypes will stick to the ridge.
Of course negative selection will keep them on what you call a "ridge"; but the pertinent question is how did it get up on that "ridge"? EL:
You are confusing the height with the horizontal distance. What is difficult for a phenotype is getting across a plain, not scaling a cliff. Let’s say that it takes 10 neutral mutations to make a functional gene, but that functional gene confers a big reproductive advantage. The difficulty is getting across the “plain” represented by those 10 functional mutations, but once there are 9 in place, it is at the foot of the “butte”, and once it gets the 10th, up it goes.
At last!! We've gotten to your notion of what happens in genomes!! I see now why you think Dawkins is so wrong. It's because you think the brunt work of getting to new function is done by neutral mutations, not positive selection. And I have a metaphor for your scenario. It's called: "Taking An Elevator Up Mount Improbable"! Or, as you Brits would say: "Taking A Lift Up Mount Improbable"! There's two ways to go here. We can assess things without assuming the need for fixation of neutral mutations. And we can assess things assuming this isn't necessary. In the first case, Kimura calculates that it takes around 3.5 million years for a SNP to become fixed for an effective population of 500,000. This has to happen 9 times given the example you're using. So that's 31.5 million yrs., simply to come up with a new functional gene. Isn't that about the time of the entire Cambrian Explosion? This is way too slow. OTOH, assuming no fixation, then any individual organism/genotype with the first neutral mutation has to stay alive. If some deleterious mutation comes along anywhere else in the genome, then the organism dies, taking with it the needed mutation. So, negative selection is there all the time, weeding out the perhaps 100's of deleterious mutations that occur during each replication. And this tightrope walk has to happen 9 times. Is this probable? Well, this looks like a "random walk" through a mine field. Next, consider this. The first neutral mutation can occur in any member of the population. But the second, the third, and so forth, mutations need to occur in the genome where the first occured since fixation isn't assumed. So, how many of these "neutral mutations" will appear in the population? We have no idea. It could be zero; it could be a hundred, a thousand, we don't know. But let's say we're dealing with a small number; let's say a hundred. Then the probability of getting the second mutation at a specific spot is 1 in 10^9. This means that if the number of members of the population having the first mutation doesn't increase, then it will take, on average, 10^7, or 10 million years to show up in those members having the first mutation. If you want to argue that during these 10 million years there will be other members of the population that will get the first mutation, then we also have to remember that during this same 10 million yrs. there will be members with the first mutation that will lose it. It's this kind of back and forth flux that works into the notion of the time needed for fixation of a mutation. In this case, a fixation time less than 10 million yrs would be preferable. So, we have a time somewhere between 3.5 million yrs. and 10 million yrs. Again, this looks to be way too slow. But let's say this is possible. Then shouldn't we see "intermediates"? That is, shouldn't we see some members of the population with 7 or 8 of the needed 9 neutral mutations? Well, where are they? Now our technology might not now be able to give us these answers, but I bet there are some studies done here and there that can even now give us hints as to whether or not we see these kinds of "intermediates". To date, I haven't heard of any such discoveries. So, here we have it: neutral drift is too slow, and positive selection is too costly (deadly--i.e., Haldane's Dilemna = genetic load). EL:
And, again, this has been extremely widely criticised. So only if the critiques can be countered, can you conclude anything from this.
I've seen some of the critcisms, and have criticized some of the criticisms. And the criticisms in general amount to this: "It doesn't take 10^20 replications to come up with just one new binding site, you can come up with quite a few!" Is this really a criticism? It sounds more like a capitulation. The other great big criticism is: "Where did Behe come up with 10^20 replications?!!!" Well, from a paper by someone working in the field. That's why I included the words "in vivo" when discussing EofE, since, hypothetically, it should only take 10^16 replications; yet, in practice, the first a.a. substitution takes 10^12, and not 10^8 replications. The 10^20 figure is one that can be calculated from the known data. When a single infection of the parasite can produce upwards of10^10 parasites in one individual, and with millions of individuals infected each year, a mimimum figure for the required number of replications has to be around 10^13 to 10^15. And that many for what: 2 a.a. substitutions!!!! Ah, yes, Climbing Mt Improbable. In what, trillion upon trillions of years? EL:
But it is one thing to say that on principle, Irreducibly Complex things cannot evolve, when a simple model shows that they can, and another to show that this allegedly IC thing, or class of things could not have evolved, in which case your conclusion is only as good as your model, as Behe and Snokes themselves point out.
Behe and Snoke's model wasn't trying to demonstrate anything whatsoever about IC. They were simply trying to accurately model the evolutionary model then current. You're "simple model" that "shows that [IC] things can [evolve]" involves propositions that have not been demonstrated, but simply assumed. OTOH, in Behe's "Edge of Evolution", nothing is assumed. It's nature at work. I'd rather base my thinking on what nature is known to have done than on undemonstrated "assumptions".PaV_{October 31, 2011
October
10
Oct
31
31
2011
11:08 PM
11
11
08
PM
PDT}

You go on affirming that. It is not true, and you have never explained why it should be so. Please, a clear and specific answer to that would be greatly appreciated.
Well, clearly I have not been clear, but I have certainly been specific! And, indeed, I have tried to explain. I'll try again, but I probably won't have time to get to this for a bit now. Thanks for your response anyway, and when I do get back, I'll try to respond your whole post.Elizabeth Liddle_{October 31, 2011
October
10
Oct
31
31
2011
05:00 PM
5
05
00
PM
PDT}

Elizabeth: Nice contribution. But, obviously, I disagree. First of all, I don't accept at all your constant reference to reproductive success as the only "function". Biochemical functions are vwry well defined, and in themselves they have nothing to do with reproductive success. An enzyme is a perfect machinery to accelerate a specific chemical reaction. It does that, and accomplishes its function, whether in a living being or in the lab. It has a perfectly definable function. Many of those biochemical functions we use out of the context of the reproduction of living beings. Indeed, protein engineering is also about the possibility to have biochemical advantages from new molecules. So yes, I do keep the biochemical function of proteins separate form their phenotypic effect on reproduction. And anyway, I am still talking of phenotype. And enztme, the protein, is phenotype. Phenotype is not only reproduction. It is much more. Another statement that is, IMO, completely wrong is the following: Except it turns out that, because of drift, those “probability barriers” are much lower than were once thought. You don't motivate that at all. It is not true. Drift does not change the probability barriers at all. Let's go through it again. My original statement is: "The process happens in a duplicated, inactivated gene. In that case, it is not clear how NS can act at all, and the variation remains by definition neutral. All the probability barriers apply." So, we are talking of a gene that is not translated, that can change freely. We are discussing the possibility that such a gene is the starting point for a different, unrelated protein domain of high complexity (more than 150 bits of dFSCI). We agree that all variation in that gene are, by definition, neutral (the gene is not translated). We agree that random genetic drift can happen. We agree, I hope, that for the new state to be reached the search must dare to go well out of the local (the state that should be reahced is unrelated at the sequence level). So I strongly state that, in this situation, all unrelated states have the same probability to be reached. Will you please explain why that should not be true, abd how drift should make the probability barrier lower? You go on affirming that. It is not true, and you have never explained why it should be so. Please, a clear and specific answer to that would be greatly appreciated. You say: An inactivated gene can still drift through the population, collecting mutations. It certainly can. But the probability of reaching a functional unrelated protein domain in that way is almost zero. The problem is not that such an achievement cannot be reached. For instance, if a designer guides the variation, it can certainly be reached, and even in a short time. On the contrary, with no guide, and by the effect of purely random variation, you need all the time of the universe to have a decent chance, even in a single case. Or more realistically, you cannot succeed even with all the time of the universe at your disposal. Well, you’ve left out viruses. Viruses are not autonomous replicators. Please, detail what your model about viruses is, and we will discuss it. But again, viruses use the already existing, complex resources of more complex beings. Even including viruses, however, the probability barriers cannot be matched. The higher replication rate of viruses can probably rise the probabilistic resources of a few orders of magnitude. But we in ID always keep tens of orders of magnitude of "buffer" in our analyses. No problem, even with viruses. you still have the burden of showing that the fitness landscape between functional domains is too wide and too featureless for probability to account for a route across it I have discussed that in detail many times. You must decide. If you are interested in that aspect (until now, you have usually avoided to discuss it), then we must go deeper into the concept of dFSCI and into the existing literature about protein function and the protein search space.gpuccio_{October 31, 2011
October
10
Oct
31
31
2011
04:15 PM
4
04
15
PM
PDT}

Elizabeth: The problem here:
Conversely, it may be that some proteins are Irreducibly Complex, in which case they are sitting on a high butte in a flat landscape. But again, that doesn’t mean they can’t be climbed, because as DrBot points out, populations can drift along horizontal ridges, and if they drift to the base of a bluff, there is nothing to stop them ascending to the top.
is that this amounts to a “just-so” story told using the language of molecular biology.
I don't know what people mean when they dismiss models as "just so" stories, PaV. I see it a lot, but I don't understand it. I'm talking about testable hypotheses, not fairy tales that do not make testable predictions. Sure, it's a metaphor, but it's a very precise metaphor, and can be translated into real productive ouputs, as DrBot will attest.
There are various parts to a protein sequence: you can have hydrophilic sections attached to hydrophobic sections, e.g., where the hydrophobic section sits in the cell membrane, while the hydrophilic section dangles outside, or inside of the cell. Likewise, there are “conserved” regions, and non-conserved regions—which easily translates into vital portions and less vital portions. Per se, neutral drift occurs in that portion of the protein sequence that is less vital. In the vital segments of the protein sequence, either positive, or negative selection is going to take place.
I thought we were talking about the origins of a novel protein domain. But either way, what matters is the effect on the phenotype. Clearly if a sequence is "vital" it will be highly conserved, because no offspring bearing a lethal mutation to it will survive to pass on that lethal mutation.
So, yes, BOTH neutral drift and NS work side-by-side.
Well, I think it's artificial to separate them, or, indeed to talk about "NS" in the same breath as "drift" anyway - "NS" is just a bias to the drift process. But as long as that's what you mean, we agree.
And even in the highly-conserved (= vital) sections, a small amount of drift can, and does occur.
Meaning what? Yet again, it isn't clear when you are talking about phenotype and when you are talking about gene! And if you are talking about a gene, then it matters whether you are talking about sexual or clonal reproduction. And also whether you are considering HGT. What do you mean when you say a "section" of a gene "drifts"? Do you mean it can accumulate minor variants and still be reproduced? Sure. And the "less vital" parts can accumulate more variation still be reproduced.
And, it is possible for a previously “neutral” mutation to assume some kind of function.
Yes indeed, by gene-gene interaction, but I'm not sure that's what you mean. Perhaps you mean that it's possible for a previously neutral phenotypic feature to become beneficial. Which is also true. But it's important to distinguish between the two, as they are quite different.
But all of this is very limited simply because the critical portions of proteins allow only the smallest amount of wiggle room (i.e., they’re “highly conserved”).
Well, of course, but that's looking from the evolved protein backwards, not from no-protein forwards. Once something has become useful, then you are right, there is less "wiggle room" - which is why gene duplication is probably an important process, and also HGT, in generating novel selectable phenotypic variance. But that's OK, because we know that both those things happen, and that there is also some "wiggle room" (in other words there are functional variants of proteins) and also that sexual reproduction introduces yet another novelty-producing mechanism, namely recombination, whereby entire domains can be diced and spliced.
What does this look like? Well, it means that if you change but ONE critical amino acid, then significant loss of function can occur; and if you change TWO critical amino acids, then death is likely to occur.
Well, no, it doesn't. But even if for some proteins it is true, then all it means is that those sequences will be highly conserved. And we know many sequences are. But being highly conserved after-the-fact is not the same as being vital before there is such a thing as the protein in the first place. Again, this is one of the many fallacies in Behe's concept of Irreducible Complexity.
IOW, there isn’t 12345; there’s, maybe, 123, with huge fall-offs associated with 3 versus 2, and even 2 versus 1.
But you are travelling the wrong way. And I'm not quite sure, even what you are referring to. If you are talking about DrBot's gene-space, the values are phenotypic fitness, and if there is a big "cliff" then that's not a problem - the phenotypes will stick to the ridge.
So you say that if you’re at the base of a “butte”, now all you have to do is scale it. Well, all you have to do to get to the next galaxy is travel faster than the speed of light. Sounds easy, doesn’t it. IOW, “scaling” the “butte” is the problem. And it’s a big problem for NS, let alone neutral drift.
No, you are misunderstanding the metaphor. The "altitude" on DrBot's fitness landscape is just that - fitness. You are confusing the height with the horizontal distance. What is difficult for a phenotype is getting across a plain, not scaling a cliff. Let's say that it takes 10 neutral mutations to make a functional gene, but that functional gene confers a big reproductive advantage. The difficulty is getting across the "plain" represented by those 10 functional mutations, but once there are 9 in place, it is at the foot of the "butte", and once it gets the 10th, up it goes. This is what I mean by it being important to get the metaphors straight, and to ensure we are talking about the same things when we use the same words! It's so easy to think you are in disagreemente, when actually you are talking about quite different things!
Behe worked with a mathematician by the name of Snoke to build a computer model of evolution that tried to simulate the model that evolutionary biologists proposed as explaining macroevolution: gene duplication, nuetral drift, selection, dah, dah, dah. Behe and Snoke constructed the model using assumptions highly favorable to Darwinian theory. Their results showed that evolutionary models were extremely limited, and that it would take huge amounts of generations to effect small changes.
Well, no. It's not my field, so I'm not in a position to evaluate it it, but if you are going to base your case on Behe & Snokes, then be aware of the limitations of their model, which the authors readily conceded in their response to the critique by Michael Lynch:
Our paper (Behe and Snoke 2004) contains one simple result. When reasonable parameters are used with our model to estimate actual time scales or population sizes for the evolution of multi-residue (MR) protein features, they are unrealistically large. This implies that the model we chose, which is restricted to point mutations and assumes intermediate states to be deleterious, isn’t a plausible evolutionary pathway. One must therefore look about for a new model. We did not rule out such a possibility; in our original article, we explicitly stated, ‘‘we should look to more complicated pathways, perhaps involving insertion, deletion, recombination, selection of intermediate states, or other mechanisms, to account for most MR protein features.’
PaV:
Based on comments he’s made here and there, I suppose he thought that his results were a little too pessimistic. So he decided to investigate further. The result was his study using the malarial parasite, and his Edge of Evolution. The parasite mutates randomly—call it a random walk, or call it neutral drift, whatever suits you—and it takes, in vivo, 10^20 replications to produce a new binding site consisting of a 2 amino acid change. These results are consistent with his computer model.
And, again, this has been extremely widely criticised. So only if the critiques can be countered, can you conclude anything from this.
So, at the end of the day, scaling “buttes” that involve changing more than ONE amino acid in the critical portion of a protein domain is almost impossible—either randomly, or even with selection in play.
Only if Behe is correct, and there have been plenty of critiques.
Certainly, to utilize simple models to help one envision difficult concepts can be worthwhile. But if the simplicity employed ends up being an over-simplification, then we’re more likely to confuse ourselves and others, than to produce anything that is helpful. As they say, “more heat, than light.”
Sure. But it is one thing to say that on principle, Irreducibly Complex things cannot evolve, when a simple model shows that they can, and another to show that this allegedly IC thing, or class of things could not have evolved, in which case your conclusion is only as good as your model, as Behe and Snokes themselves point out. And in any case model that aims to show that something "could not" happen is a very weak model, because at best you can conclude that your model is wrong, not that no comparable model is correct.Elizabeth Liddle_{October 31, 2011
October
10
Oct
31
31
2011
03:24 PM
3
03
24
PM
PDT}

Elizabeth: The problem here:
Conversely, it may be that some proteins are Irreducibly Complex, in which case they are sitting on a high butte in a flat landscape. But again, that doesn’t mean they can’t be climbed, because as DrBot points out, populations can drift along horizontal ridges, and if they drift to the base of a bluff, there is nothing to stop them ascending to the top.
is that this amounts to a "just-so" story told using the language of molecular biology. There are various parts to a protein sequence: you can have hydrophilic sections attached to hydrophobic sections, e.g., where the hydrophobic section sits in the cell membrane, while the hydrophilic section dangles outside, or inside of the cell. Likewise, there are "conserved" regions, and non-conserved regions---which easily translates into vital portions and less vital portions. Per se, neutral drift occurs in that portion of the protein sequence that is less vital. In the vital segments of the protein sequence, either positive, or negative selection is going to take place. So, yes, BOTH neutral drift and NS work side-by-side. And even in the highly-conserved (= vital) sections, a small amount of drift can, and does occur. And, it is possible for a previously "neutral" mutation to assume some kind of function. But all of this is very limited simply because the critical portions of proteins allow only the smallest amount of wiggle room (i.e., they're "highly conserved"). What does this look like? Well, it means that if you change but ONE critical amino acid, then significant loss of function can occur; and if you change TWO critical amino acids, then death is likely to occur. IOW, there isn't 12345; there's, maybe, 123, with huge fall-offs associated with 3 versus 2, and even 2 versus 1. So you say that if you're at the base of a "butte", now all you have to do is scale it. Well, all you have to do to get to the next galaxy is travel faster than the speed of light. Sounds easy, doesn't it. IOW, "scaling" the "butte" is the problem. And it's a big problem for NS, let alone neutral drift. Behe worked with a mathematician by the name of Snoke to build a computer model of evolution that tried to simulate the model that evolutionary biologists proposed as explaining macroevolution: gene duplication, nuetral drift, selection, dah, dah, dah. Behe and Snoke constructed the model using assumptions highly favorable to Darwinian theory. Their results showed that evolutionary models were extremely limited, and that it would take huge amounts of generations to effect small changes. Based on comments he's made here and there, I suppose he thought that his results were a little too pessimistic. So he decided to investigate further. The result was his study using the malarial parasite, and his Edge of Evolution. The parasite mutates randomly---call it a random walk, or call it neutral drift, whatever suits you---and it takes, in vivo, 10^20 replications to produce a new binding site consisting of a 2 amino acid change. These results are consistent with his computer model. So, at the end of the day, scaling "buttes" that involve changing more than ONE amino acid in the critical portion of a protein domain is almost impossible---either randomly, or even with selection in play. Certainly, to utilize simple models to help one envision difficult concepts can be worthwhile. But if the simplicity employed ends up being an over-simplification, then we're more likely to confuse ourselves and others, than to produce anything that is helpful. As they say, "more heat, than light."PaV_{October 31, 2011
October
10
Oct
31
31
2011
02:07 PM
2
02
07
PM
PDT}

Hi there, gpuccio!
Guys (Elizabeth, DrBot): First of all I want to thank you both for your intelligent contributions to this discussion. You are good interlocutors indeed.
And thank you to you too :)
While we catch our breath, I would like to add a couple of comments. Much confusion in your answers is, IMO, created by the fact that you refer to models and context that are completely different from what I am discussing (the origin of protein domains).
Yes, I think so. I think a lot the disagreements on this site arise because we are "divided by a common language", or by what linguists call "false friends" - we think we are talking about the same thing, but we are not! And one of the problems I keep trying to draw attention to is that it is often not clear whether we are talking about genotypes or phenotypes. It is the phenotype that has "fitness", not the genotype. And in sexually reproducing organisms, the genes are independent, which is, again, important when discussing "drift". In what follows you are clearly talking about genes and genotypes, and it is important to bear this in mind when trying to figure out where you are disagreeing with DrBot, who is talking about the fitness of phenotypes within "gene space".
The differences are many, but I would like to list briefly the most important: a) A protein domain is a functional unit, and a complex functional unit. The mean length of a protein domain is more than 100 AAs, but many are quite longer. A protein domain has a very specific biochemical function. This function is what must be considered first when we discuss the origin of domains. A protein domain, being a functional unit, cannot be decomposed in simpler functional units. Most protein domains are certainly well beyond conventional limit of 150 bits for the dFSCI in biological systems.
And here I would disagree! We cannot consider "function" outside the context of the phenotype, and a gene for a protein doesn't have a function unless that gene is expressed, and whether its function is beneficial, neutral, or deleteriouis for the phenotype will depend on the conditions under which it is expressed. And while a protein domain may not exist as such until a genetic sequence coding for it is generated, that doesn't mean that a part-sequence can't be a precursor of the whole. It's just that, presumably, those part-sequences will be neutral with respect to the fortunes of the phenotype - so represent "flat" areas of the fitness landscape. If indeed they are not transferred, wholesale from viral sources, which is one important hypothesis.
b) The infamous concept of “fitness function” is not good at all when speaking of protein domains. There is no simple or continuos fitness function that can be applied to that context.
Yes, there is. Ignoring viruses for a minute, let's take a sequence of base pairs that doesn't code for anything, and a sequence of basepairs that codes for the simplest protein domain. And let's assume that the protein, if it is made, does something useful for the organism. Any changes to the original sequence aren't going to do much for the phenotype, so the fitness landscape in that "gene space" (or let's call it "sequence space, as it isn't really even a gene yet) is flat. Any change will result in no change in phenotypic fitness. Then it hits a protein domain. Bingo - leap in fitness. That's a continuous function - sure it's got a big step in it, but that doesn't matter. What does matter is how big the plain is at the foot, which may be pretty big. Which is why the potential role of viruses is so intriguing.
As we are discussing NS, the relationship between the specific biocemical function of a protein and the reproductive function can be very complex and indirect. That’s why, first of all, we must look at the biochemical function if the protein. That function must be present. It can be measured independently form the reproductive function. Therefore, we, as intelligent observers, can define and measure the biochemical function in precise and direct ways.
Well, sure, but that's after-the-fact. Once we have a protein that clearly does something to enhance reproductive success, however indirectly, then we have, by definition, a function that is "selected" (or, once present, highly conserved). What its function is may be of great interest, but from the point of view of biology, if it doesn't help the organism reproduce, it isn't functional at all, it's a malfunction (however much fun it is for the organism).
But NS can “see” the single protein, both positively and negatively, only in the measure that the specific biochemical function of that protein is fundamental for reproductive fitness.
And I think you are getting into a muddle by trying to split function from reproductive success. NS doesn't "see" anything, of course (and your scare quotes indicate that you know this, but perhaps you don't know it quite deeply enough!). What we have is a protein that may or may not be expressed in a way that enhances the organism's chance of reproductive success. If it doesn't, its sequence is less likely to be conserved. If it does, it is more likely. And clearly if it isn't produced at all, because the sequence doesn't code for anything yet, then it won't do anything, beneficial or otherwise, for the organism, except, possibly be a slight metabolic luxury.
So, indeed, a mutation can be negative, or positive, in the biochemical sense, and still be invisible to NS.
No. This makes no sense at all. You can't consider "function" without considering the phenotype, and if a protein is made, or not made, and can do something, or not do something, or be expressed appropriately, or inappropriately for the phenotype, that cannot be "invisible to NS" because all "NS" is is differential reproductive success. And if the mutation has phenotypic effects and those phenotypic effects affect reproductive success, then that is NS. And if the new protein does nothing to affect the phenotype's reproductive success as compared with a peer without the genetic ability to make the protein, then that protein doesn't have a "function". It's just decoration.
c) Protein domains are isolated islands of functionality. They have different structures and different functions.
Well, they may be "islands" in the sense that on a fitness landscape they are buttes in a wide plain, but that doesn't mean that they can't be reached, because, as DrBot says, drift can carry populations across wide plains.
Most of the experience in neutral evolution, so dear to Elizabeth, is derived form observation of the neutral evolution inside families. As I have already said in my previous post, that kind of evolution, although very important to assess relationships between species, doe not affect the function, or at most (but it can be debated, and is debated by Axe in his paper) could be repsonsible of small tweaks at the active site level, with changes of function but preservation of the general structure.
Actually, it's derived from fairly straightforward models! In which quite complex functional sequences can evolve, despite many necessary precursor steps being neutral with regard to the reproductive success of the phenotype. I do think you are being mislead by trying to consider "function" outside the context of the phenotype, and, specifically, outside the context of the fitness of the phenotype - its reproductive success.
That kind of model does not apply to the generation of new domains, where a whole and completely new sequence of more than 100 AAs arises.
Yes it does, because there is no reason to think that the sequence had no incremental precursors, even if we assume that they were not even semi-functional (which may not be the case). Although there is also the possibility that they evolved within viruses (which evolve much more rapidly) and actually did plonk themselves into the genome of cells. There's at least some evidence IIRC that HGT and domain genesis ab initio, as it were, both contribute to the evolution of new domains. I'll try to dig out the paper I read.
For that kind of event, only three models can be conceived:
*sucks teeth* Rash assertion! Never underestimate the number of models that can be conceived! What matters is how many make testable predictions that are subsequently supported!
1) An existing, functional, transcirbed and translated gene gradually changes to a new gene. That is full of problems: fisrt of all the original function is rapidly lost, and it is not clear how NS would act, given the two different functions of A and B.
An existing useful gene will tend to be highly conserved, so if by "functional" you mean "contributes to evolutionary success" yes indeed. That is in accord with DrBot's model of the "ridge".
2) The process happens in a duplicated, inactivated gene. In that case, it is not clear how NS can act at all, and the variation remains by definition neutral. All the probability barriers apply.
Except it turns out that, because of drift, those "probability barriers" are much lower than were once thought. An inactivated gene can still drift through the population, collecting mutations. This is why Behe's IC notion just isn't the barrier to evolution he thinks it is.
3) The process happens in a segment of non coding DNA, possibly through transposon activity. All the probability barriers still apply, and NS cannot act (at least, not until a functional transcribed, translated protein does arise.
Sure, but see above. Same possibility applies.
My favourite model is definitely #3. That is the best model for design. And transposons are, at present, the best candidates for guided variation.
Well, you've left out viruses. And you still have the burden of showing that the fitness landscape between functional domains is too wide and too featureless for probability to account for a route across it. Anyway, nice to talk to you again, will drop by in a bit in case you respond (I feel I'm departing UD like the Cheshire Cat, so I'll leave you with a :D)Elizabeth Liddle_{October 31, 2011
October
10
Oct
31
31
2011
01:56 PM
1
01
56
PM
PDT}

I don't respond to "arguments" that I don't understand, or that just don't exist. Be more clear, and we'll see...gpuccio_{October 31, 2011
October
10
Oct
31
31
2011
11:48 AM
11
11
48
AM
PDT}

PaV: I absolutely agree. The role of transposons in primates is obvious, but it probably extends backwards. Another example is the rather sudden "origin" of the specific immune system in jawed fish, which is also attributed to transposon activity (with the appearance of the fundamental proteins RAG1 and RAG2: and no kidding here, human RAG1 is 1043 AAs long!)gpuccio_{October 31, 2011
October
10
Oct
31
31
2011
11:47 AM
11
11
47
AM
PDT}

DrBot: I respect your flu (I had one last week, and I am still not fine). So, brief comments: So, the probabilities depend on the relative co-ordinates of the two locations in configuration space. Exactly my point. The configuration space is the space of sequences. Do you agree with that? I ask, because that seems to be a very simple point that darwinists often forget. Variation acts on the sequence. More the sequence is different, more the proteins are far away. So, two sequences that have less than 10% identities are as far as possible in that space. Basic protein domains share less than 10% sequence. They are as far as possible in the space. The space has to be traversed, to go from one to the other. Basically NS prevents the search ever getting to a point more than waist deep in the sea – so the deep sea, the vast area of ‘dead’ genomes, never gets explored. Negative selection presents a barrier that locks out these parts of the search space and effectively makes the searchable space much smaller that the whole of the configuration space. Exactly my point. Negative selection is exactly responsible for keeping the search inside the functional island. That's why existing proteins remain functional, and retain structure and function, thorughout natural history, even if their sequence changes because of neutral variation. That's why we have so many different myoglobins, so many different aminoacyl tRNA transerases, that are essentially the same protein, although their sequences are sometimes different enough. That is calling "traversing the functional island". Please, look at the followinf paper: "Sequence space and the ongoing expansion of the protein universe" http://www.nature.com/nature/journal/v465/n7300/full/nature09105.html summarized here (the paper is not free): http://www.lucasbrouwers.nl/blog/2010/05/the-big-bang-of-the-protein-universe/ As the paper states, may proteins have not even traversed fully their own functional island. But again, that is what happens when A is functional and necessary: A, very simply, is preserved. Its sequence can change somewhat, but it is neutral change. IOWs, A does not go anywhere. It remains in its functional island. If that were all that happens, new proteins would never appear. Or at least, not new protein domains (As I have discussed, some microevolution, or let's say "intermediate evolution", implemented by changes of 1 - 8 AAs, could still happen, and we can discuss where really is the "edge" of unguided functional variation. Personally, I would stick to Behe's 2 or 3 AAs). But that is not all that happens. New basic protein domains, as I have argued, have been appearing throughout natural history. Even in humans there are probably a few. So, where do they come from? Again, the "duplicated gene" and the "non coding DNA" scenarios seem the best. They both have three things in common: a) NS, both negative and positive, seems to have no role, or only a really minor role b) Neutral variation can easily happen, and can easily reach any possible target c) The whole dFSCI barrier applies. This ratio of searchable space to search space is critical IMO. Sure. That is the main point in dFSCI calculation. The Durston method remains the best way to do that. This is why you have to consider both the search space topology Sure. And the search space topology is extremely simple. The greater the different between two sequences, the greater the distance. Very simple, isn't it? There is not much to be considered. Why is it so? Simply because variation is applied to the primary sequence. Variation happens randomly at DNA level. It has nothing to do with the protein. It has nothing to do with the phenotype. It's only NS that "sees" the phenotype, not RV. Variation is a variation of sequence. The search space is a sequence space. and the nature of the search (i.e. localised/stepwise or random sampling) The nature of the search is clear. Neutral search that does not affect the function, or at least the folding, is localised (traversing the functional island). In no way it can explain new protein domains. Free neutral search, like in duplicated genes or non coding DNA, can search the whole space. It just needs time to get away from the initial state (not too much, indeed). Jump is possible, but limited to some forms of variation (such as frameshift mutation). The distribution of probabilities for achieving any particular state from a given point and in a given time are highly dependant on how your search proceeds and the shape of the landscape you are traversing. Sure. And that's exactly what I have tried to analyze in detail.gpuccio_{October 31, 2011
October
10
Oct
31
31
2011
11:37 AM
11
11
37
AM
PDT}

gpuccio: My favourite model is definitely #3. That is the best model for design. And transposons are, at present, the best candidates for guided variation. This certainly applies to primates. And is probably the best suggestion of inherent design in the genome.PaV_{October 31, 2011
October
10
Oct
31
31
2011
09:26 AM
9
09
26
AM
PDT}

Yes, I meant the vertical slopes of fitness. It would really be nice to trace that. Thanks anyway.Eugene S_{October 31, 2011
October
10
Oct
31
31
2011
08:38 AM
8
08
38
AM
PDT}

I have no time to do anything on this topic until this afternoon, which, per my location, is about 8 hours hence. Til then.PaV_{October 31, 2011
October
10
Oct
31
31
2011
08:21 AM
8
08
21
AM
PDT}

Eugene S: I suppose you're talking about the fitness landscape being so vertical? If so, I regret not bookmarking it, and so I simply can't remember where it was. It was in a paper that was considering something else; and I think that's the only reason we see it in print. I, too, have long suspected--really, known--that this is what they must look like. But the EvBiol don't want to look these things square in the eye. Yet, this is what fitness landscapes really look like. As to Stuart Kauffman: yes, he's gone on to greener pastures. I suspect he might not come out and say that the rugged landscape approach has no merit; but the fact that he's moved on says a lot. Should I come across the paper again, I'll try to reference it here. It's paper that's come out within the last two years, IIRC. But that's not narrowing it down too much, is it? I have to run . . .PaV_{October 31, 2011
October
10
Oct
31
31
2011
08:18 AM
8
08
18
AM
PDT}

Flu? Drink some emegence-c and some cell power- should be available at your local health-food store Three times a day- and if you continue this you won't get the flu again....Joseph_{October 31, 2011
October
10
Oct
31
31
2011
07:12 AM
7
07
12
AM
PDT}

I think it is a poor analogy to support the idea of biological evolution as the languages do not reproduce and because intelligent agency is directly involved in language formation. In any case it is no better than any artefact.Eugene S_{October 31, 2011
October
10
Oct
31
31
2011
07:02 AM
7
07
02
AM
PDT}

Ok, I appear to have caught a bout of flu and now my brain feels like it is full of toxic mud, so rather than providing a long response I'll just reply to a few key points for now - then have some pills and go to bed :( The first general point is that within a search space that is being traversed in small steps, in the way evolution does, rather than random bounds, in the way a random search does, the probabilities of getting from point A to point B scale with distance - this should be obvious really. The 'search' proceeds by taking relatively small steps in the local configuration space, resulting in some kind of fitness feedback (a change in reproductive success). Because of this within a given time period your probability of reaching a state diminishes with the distance of that state from your starting point. So, the probabilities depend on the relative co-ordinates of the two locations in configuration space. Second - because NS will act on ANY genome configuration that enhances or diminishes reproductive success, some locations in the search space are less likely to be traversed. If you think of in in terms of the islands metaphor: 1 Above seal level is positive function (enhances reproductive success) 2 At sea level is zero function (but not detrimental - just 'does nothing') 3 Below sea level is detrimental (reduces reproductive success) 4 Neutrality is ANY flat path or ridge (not necessarily the at sea level) 5 And finally - go too deep (up to your neck) and you reach non viability (no reproduction at all) Basically NS prevents the search ever getting to a point more than waist deep in the sea - so the deep sea, the vast area of 'dead' genomes, never gets explored. Negative selection presents a barrier that locks out these parts of the search space and effectively makes the searchable space much smaller that the whole of the configuration space. This ratio of searchable space to search space is critical IMO. This is why you have to consider both the search space topology, and the nature of the search (i.e. localised/stepwise or random sampling) if you want to calculate meaningful probabilities. The distribution of probabilities for achieving any particular state from a given point and in a given time are highly dependant on how your search proceeds and the shape of the landscape you are traversing.DrBot_{October 31, 2011
October
10
Oct
31
31
2011
07:01 AM
7
07
01
AM
PDT}

We know from actual experiment that it is possible to select minimally functional sequences from a random library of sequences. Now which is the more rational assumption: quintillions of bacteria produce a minimally functional mutation every few million years, or that an invisible sky pixie drops in every few million years an poofs a sequence into existence. There's no proof either way, but you can accept a process that has been observed, or one which hasn't. I don't believe you ever responded as to why you think speciation of protein coding sequences and the resulting isolation is a problem. I believe I asked about the Basque language and whether it was poofed into existence.Petrushka_{October 31, 2011
October
10
Oct
31
31
2011
06:36 AM
6
06
36
AM
PDT}

Guys (Elizabeth, DrBot): First of all I want to thank you both for your intelligent contributions to this discussion. You are good interlocutors indeed. While we catch our breath, I would like to add a couple of comments. Much confusion in your answers is, IMO, created by the fact that you refer to models and context that are completely different from what I am discussing (the origin of protein domains). The differences are many, but I would like to list briefly the most important: a) A protein domain is a functional unit, and a complex functional unit. The mean length of a protein domain is more than 100 AAs, but many are quite longer. A protein domain has a very specific biochemical function. This function is what must be considered first when we discuss the origin of domains. A protein domain, being a functional unit, cannot be decomposed in simpler functional units. Most protein domains are certainly well beyond conventional limit of 150 bits for the dFSCI in biological systems. b) The infamous concept of "fitness function" is not good at all when speaking of protein domains. There is no simple or continuos fitness function that can be applied to that context. As we are discussing NS, the relationship between the specific biocemical function of a protein and the reproductive function can be very complex and indirect. That's why, first of all, we must look at the biochemical function if the protein. That function must be present. It can be measured independently form the reproductive function. Therefore, we, as intelligent observers, can define and measure the biochemical function in precise and direct ways. But NS can "see" the single protein, both positively and negatively, only in the measure that the specific biochemical function of that protein is fundamental for reproductive fitness. So, indeed, a mutation can be negative, or positive, in the biochemical sense, and still be invisible to NS. c) Protein domains are isolated islands of functionality. They have different structures and different functions. Most of the experience in neutral evolution, so dear to Elizabeth, is derived form observation of the neutral evolution inside families. As I have already said in my previous post, that kind of evolution, although very important to assess relationships between species, doe not affect the function, or at most (but it can be debated, and is debated by Axe in his paper) could be repsonsible of small tweaks at the active site level, with changes of function but preservation of the general structure. That kind of model does not apply to the generation of new domains, where a whole and completely new sequence of more than 100 AAs arises. For that kind of event, only three models can be conceived: 1) An existing, functional, transcirbed and translated gene gradually changes to a new gene. That is full of problems: fisrt of all the original function is rapidly lost, and it is not clear how NS would act, given the two different functions of A and B. 2) The process happens in a duplicated, inactivated gene. In that case, it is not clear how NS can act at all, and the variation remains by definition neutral. All the probability barriers apply. 3) The process happens in a segment of non coding DNA, possibly through transposon activity. All the probability barriers still apply, and NS cannot act (at least, not until a functional transcribed, translated protein does arise. My favourite model is definitely #3. That is the best model for design. And transposons are, at present, the best candidates for guided variation.gpuccio_{October 31, 2011
October
10
Oct
31
31
2011
06:12 AM
6
06
12
AM
PDT}

PaV, Extremely interesting. I have long been sure that this would be the case, but I did not have concrete figures. I was looking for that sort of computations. Somewhere I read that somebody came up with this sort of estimates showing the practical impossibility of the evolution of the human eye. But I could not find the figures. Could you point to the paper(s) you talked about! Thanks. Maybe it's mean of me but I am glad to hear Stuart Kauffman gave up on it :)Eugene S_{October 30, 2011
October
10
Oct
30
30
2011
01:29 PM
1
01
29
PM
PDT}

Just briefly, whilst I have a few minutes to myself :) My toy example was not intended as a model of biology - and I didn't present it as such. It is designed for the sole purpose of explaining how the topology of a fitness (or inversely, and error) landscape influences the probabilities of achieving certain states during a 'search'. This concept applies to any search system and understanding in is IMO a vital part of understanding search systems in general. PaV - when you say it is not a good thing for comp-sci students to learn about because it is not an accurate model of biology you are missing two vital points. Firstly, it is a high level concept common to any search problem (including biology), and my example is designed to explain that concept in as simple a way as possible. Secondly - not all evolutionary algorithms are designed to model biology, they are widely used in engineering as tools for solving design problems, and again, understanding how fitness landscape topology affects search is vital if you want to deploy these types of algorithms effectively in real world settings - that is why many people study neutrality in fitness landscapes in detail, because it matters! I'll try and post a more substantive response tomorrow afternoon - once I've done some of the things I am supposed to do in order to earn a living ;)DrBot_{October 30, 2011
October
10
Oct
30
30
2011
11:54 AM
11
11
54
AM
PDT}

PaV, would you take a look at my comment above? I think there is a mismatch between what Dr Bot is talking about and what you are talking about. In DrBot's toy example, the fitness landscape is in gene space and what "moves" in gene space is the phenotype. It is the phenotype that changes in fitness as it moves through gene space, not the gene, which simply exists in various states (represented by the two axes), which we call alleles. So, along the horizontal and vertical axes, each row/column represents a different allele which may give rise to a slightly different protein, or even a slightly more or less efficiently produced protein (there's some evidence that even "silent" mutations have phenotypic effects). So when you say:
(1) your putative “fitness landscape” looks nothing like the real world—-real-world, actual, fitness landscapes are extremely steep. I recently looked at a paper that couldn’t have been more than two years old. In it was a graph that was comparing fitness levels to some other parameter. It was basically a fitness landscape along the horizontal portion. For, IIRC, this protein, or perhaps a protein cluster, the fitness landscape was a straight-up vertical line! Stuart Kaufman gave up on the idea of these kinds of landscapes because they were too steep to climb. So, your example—with 3s, 4s, and 5s—has nothing to do with biological reality, I’m afraid
You are only looking at a single slice through DrBot's gene space. So to take your example - if, along a single dimension represented by alleles of a single gene, a neighbouring allele confers a large increase in fitness on the phenotype, then the population will tend to "move" to the top of that "cliff" (because individuals bearing that allele will out-replicate their peers, by definition). However, if, along that single dimension there is a local maximum, then it will be harder for the population to move from that local peak. Until we consider other dimensions (two, in DrBot's toy example). For example, let's take this one: 22411 14412 14321 14421 13521 And let's say the phenotype starts at row 1, column 1, at fitness level 2. It does not move down the column, because that would mean going to a lower fitness level, and selection keeps it on its ridge. So it can drift to row 1 column 2 (a neighbouring allele of the horizontal gene) then leap up to row 1 column 3 (another neighbouring allele of the horizontal gene). Now, if we consider only that gene, it is stuck. There is a 5 on the fitness landscape, but it's on the bottom row. However, if we consider the vertical gene, there is a route to the 5 that does not involve going down hill. It's not a straight route, but it exists. And the more dimensions you add (and there will be one for every gene, and that's before we consider new genes, and of course each dimension will be much longer than five alleles) the richer the network of ridges between peaks. That is DrBot's point - that high dimensioned (even two dimensional) fitness space offers orders of magnitude more possibility from getting from one peak to another. As for some "slopes" being "too steep to climb" :) The kind of fitness landscape that is difficult for evolutionary processes to traverse are not landscapes with steep slopes, but landscapes in which there are too many plateaux and canyons between the bases of the slopes. Once you are at the base of a slope, it doesn't matter how steep it is. The issue is how you get to the base of the slope (or the edge of the cliff if you turn it upside down).
(2) While saying that moving from 4 to 3 is a deleterious mutation, you’re basically covering up, again, perhaps unintentionally, the fact that we’re really not talking about a fitness level that goes from 4 to 3, but from 4 to (-4). This is consistent with the point I’ve made in (1) above. That is, in the portions of any protein domain that codes for functionality, one amino acid substitution is sufficient for near loss of function, or even complete loss of function. You can easily go from 100% function to .001% of function. This isn’t 4 to 3.
Well again, you seem to be confusing phenotype with genotype. Fitness is a property of the phenotype, not the genotype. Now it may be that some proteins are absolutely necessary for viability, and any mutation to the coding for them is disastrous. If that is the case, then clearly along dimensions orthogonal to that represented by that gene there will be a ridge. But that doesn't mean the phenotype is stuck at a fitness maximum because there are other dimensions. It just means that that gene will be very highly conserved. Other genes are far more robust, which is why there are so many perfectly viable alleles in most populations, with near-neutral phenotypic effects. Conversely, it may be that some proteins are Irreducibly Complex, in which case they are sitting on a high butte in a flat landscape. But again, that doesn't mean they can't be climbed, because as DrBot points out, populations can drift along horizontal ridges, and if they drift to the base of a bluff, there is nothing to stop them ascending to the top. As for the "fixation" issue - sure, once an allele is fixed, it can't drift out again. But that isn't what matters. What matters is how many bearers of the allele there are. The more there are, the more opportunities there are for a subsequent mutation to occur that renders it beneficial. And the larger the population, the larger this number can be. It doesn't matter if fixation is less likely, so is extinction, once the numbers are substantial, as Kimura's equations show.Elizabeth Liddle_{October 30, 2011
October
10
Oct
30
30
2011
08:44 AM
8
08
44
AM
PDT}

DrBot: May I please respond? Let me get to the point. The model you propose looks nothing like what evolution in the real world looks like. While this example is apparently used to instruct students in the use of evolutionary algorithms, sadly, what is being programmed is an unrecognizable facsimile of what actually takes place. It amounts to a semi-sophisticated, computer-generated, "just-so" story. I will only make two observations here to make my point: (1) your putative "fitness landscape" looks nothing like the real world----real-world, actual, fitness landscapes are extremely steep. I recently looked at a paper that couldn't have been more than two years old. In it was a graph that was comparing fitness levels to some other parameter. It was basically a fitness landscape along the horizontal portion. For, IIRC, this protein, or perhaps a protein cluster, the fitness landscape was a straight-up vertical line! Stuart Kaufman gave up on the idea of these kinds of landscapes because they were too steep to climb. So, your example---with 3s, 4s, and 5s---has nothing to do with biological reality, I'm afraid (2) While saying that moving from 4 to 3 is a deleterious mutation, you're basically covering up, again, perhaps unintentionally, the fact that we're really not talking about a fitness level that goes from 4 to 3, but from 4 to (-4). This is consistent with the point I've made in (1) above. That is, in the portions of any protein domain that codes for functionality, one amino acid substitution is sufficient for near loss of function, or even complete loss of function. You can easily go from 100% function to .001% of function. This isn't 4 to 3. Because of (1) and (2), you need to have very specific amino acid substitutions that produce beneficial effects. This obviously has to occur in the functional portion of the protein domain. But this is exactly the portion of the protein sequence that is the "steepest", and the most likely portion of protein sequence where negative selection will occur. The effect of all of this will be that any "beneficial" mutation in an individual---assuming that several amino acid substitutions must occur for a new protein function to arise---can be easily overcome by a "deleterious" mutation before the next "beneficial" mutation arises, and, thus, the whole leading to the loss of that lineage. This is partly why it takes 1/4N generations for a neutral mutation to become fixed. Now, if the population is small, then getting the "neutral" mutation at just the right place will take a long time. OTOH, if the population is large, then it takes a long time to "fix". And, if it's six, or seven, amino acid substitutions that are needed to move from one protein domain to another, this represents a long period of time. I've done some calculations in the past, and I think the optimum population size is around 500,000; which is just the population size Kimura used in the example I cited above. So, for six needed, mutations, based on a yearly reproductive cycle (Kimura's numbers reflect a two-year reproductive cycle), this amounts to 6 x 3.5 million yrs. = 21 million years for a single protein to develop a new function via neutral drift (negative selection is presupposed in all of this). So, by not including either the number of mutations that will be required, or the problem of fixation*, or the amount of time either the generation or fixation of the mutation will take, you're not anywhere near simulating a real biological example of evolution. Isn't this really the problem with most EAs? Of, if it reasonably simulates actual real-world evolutionary conditions, then the amount of real information/function it produces is negligible (and always subject to other assumptions which can sort of sneak function/information in.) ______________________________________________________ * The "problem of fixation" is this: unless the new a.a. substitution is "fixed" in the population, then the individual in whom the first a.a. substitution took place---we're here assuming this is taking place via neutral drift---has no selective advantage, and so there's no certainty whatsoever that this potentially helpful neutral mutation (PHNM)will spread. So, there will be only a small number of individuals in the population having this PHNM. Now what is the likelihood that the next PHNM will take place in these same, few individuals? Some numbers: N = 100,000; n = number of individuals with the first PHNM = PHNM1 = 100 = 0.1% of the population; genome length = 10^9 bp; mutation rate = 10^-7/generation; yearly reproductive cycle. Then, for PHNM2 to arise, we have: 100 x 10^-7 mutations/ generation/10^9 bp = 10^4 bp substitutions per generation. We need to cover the entire genome to be sure the PHNM2 arises. So, 10^5 generations will be needed = 100,000 years. However, in this period of 100,000 years, among these 100 individuals, 10^9 mutations will have arisen, of which, only one is the one we're looking for. What is the likelihood, then, that these 100 individuals would have survived this period given that, on average, every location on the genome has had a substitution? Given that negative selection is always at hand, not very likely. If, however, the substitution is fixed, then the PHNM2 will arise much faster, and with much fewer deleterious mutations to contend with. But now we're back to 3.5 million yrs. And so it goes.PaV_{October 30, 2011
October
10
Oct
30
30
2011
05:24 AM
5
05
24
AM
PDT}

Just another note: drift does not “enrich” anything. If a neutral mutation es expanded by drift, other neutral mutations are lost in that expansion. There is no enrichment.
By "enrich" I simply meant that neutral mutations are constantly present in substantial numbers in the population, representing potentially useful variance. Yes, some will be lost as others are gained, but that means there is homeostasis as regards variance, and the larger the population, the larger the variance.
Drift is absolutely useless, or at least it does nothing different from what other kinds of RV do: it randomly changes the existing sequences.
We seem to be talking about radically different things. I am not talking about anything that "changes the existing sequences". I am simply talking about the propagation of novel genetic variants through a population.
We cannot even say that drifts accelerates the rate of variation. First of all it is a rare event. Other forms of variation are much more common.
I'm at a loss, gpuccio. Could you say how you are defining "drift"? What is a "rare event"? It happens in every generation of a population! In other words, in every generation, neutral alleles will change in prevalence. Some will drop out completely. Others will be represented for the first time. But that drift process never stops. How could it?
Secondly, we know well that too much variation is worse than too little. That is well known, I believe, in population genetics. You cannot increase your chances of reaching new functional targets only by increasing the rate of variation, be it by drift, or by any other means.
You seem to be talking about mutation rate, not drift. Sure, there is probably an optimal mutation rate. But isn't what we are talking about.
Just an example of how drift can be in the same way useful or deleterious, so that its global effect is nil.
I'm really not following you. Do you mean mutation or what? I'm not sure what the antecedent of "its" is in the above sentence. "Drift" is used to describe the change in prevalance of alleles from generation to generation when the alleles are near-neutral. But because neutral variants can so "drift", and because neutral mutations are very common, it means that there will always be plenty of potentially useful variance in a population should the environment change.
Let’s say that to reach B you need serine at position 126. Well, you get lucky enough and get that mutation in A. It is not amlified, but luckily it is neutral, so it is not lost too. It is there, ready to contribute to the final result (provided that other lucky events accumulate). Now, if by further luch that neutral mutation is expanded by drift, we will certainly be nearer to the goal.
Well, yes, if that mutation occurs it may well increase in prevalence purely by drift, increasing the probability that the next "lucky" event will occur in at least one bearer.
But in the population there are many other neutral mutations. After all, as you say, neutral mutations don’t tend to favour what works better.
Sure. But that doesn't matter. What matters is that there are a substantial number of instances of the A mutation in the population, because the more there are, the more opportunities there are for the a beneficial mutation to occur in one of them.
Now, let’s say that for once you are not lucky, and the neutral mutation that is expanded by drift is one, in another individual, that has glycine at position 126.
I am really not following you. Lots of neutral mutations can become more prevalent by drift. Or are you talking about asexually reproducting populations?
Now, that is a neutral mutation for A, but it is not neutral at all for the future target, B. Indeed, let’s say that with glycine at 126, B will never be functional. Now, if the neutral glycine is expanded by drift, instead of the neutral serine (they have the same probability, after all), what has happened? You have lost the favourable neutral mutation that had already accumulated. That is the truth for neutral mutations and genetic drift. They don’t change the explanatory scenario. The only place where they are “useful” are the wrong arguments of darwinists.
tbh, what you are saying seems to me extremely muddled. I'm not sure if it's a jargon problem, or what, but I can't even parse what you are saying. Are you talking about sexually reproducting populations? And by "amplified" do you mean, is reproduced in lots of individuals? Or something else? Let me give a toy example, assuming a sexually reproducing population of 5000: Let's say that if mutation B occurs in the same individual as mutation A, the effect is beneficial for the phenotype, but that neither A nor B alone confer any reproductive benefit, nor disadvantage Let's say A occurs, and happens to an individual who happens to produce several viable offspring, all of which go on to have further offspring, resulting, over a few generations, of 1000 individuals bearing A. Now B occurs in an individual without A. This individual also reproduces successfully. Eventually, one if its B bearing descendents mates with an A bearer, producing offspring with A and B. These then produce lots of offspring, increasing the number of A and B alleles in the population. Whenever they occur in the same individual, that individual has lots of extra offspring, and eventually both A and B become "fixed" and every member of the population has both, benefiting from the resulting phenotype. Now let's take a cloning population. Here, B has to occur de novo in an individual that already has A. But the more A individuals that exist the more chance there is for B to occur in at least one of them. So again, providing A doesn't do any harm, there's a good chance that there will eventually be lots of individuals with A, each additional individual increasing the probability that one of them will end up with B mutation. So either way, the fact that neutral mutations can drift increases the number of opportunities for a second mutation that converts the first into a benefit to occur. No?Elizabeth Liddle_{October 29, 2011
October
10
Oct
29
29
2011
04:33 PM
4
04
33
PM
PDT}

I'm chastened. And amused.PaV_{October 29, 2011
October
10
Oct
29
29
2011
03:49 PM
3
03
49
PM
PDT}

PaV:
Lizzie: I’ll assume you’ll look at my response at 47.2.1.1.7 above.
Yes indeed :) It is not my intention to evade responding - I just need to spend more time at my own site, and try not to start more hares here!
Lizzie: Sorry for taking so long to reply. I must admit that with this new format, and with the sheer number of posts taking place each day, it’s very hard to keep track of where I’ve left an entry. (Hope somebody’s taking note) But, I’v also been away from a computer for a couple of days.
Yes, that's one reason I would welcome people here dropping by at my own site, where things move a little more slowly! Not trying to poach, just trying to provide a more leisurely environment for longer conversations: http://theskepticalzone.com/wp/ Would be delighted if you dropped by :) OK:

I don’t know what you call it, PaV, and I don’t care very much; the important thing is that it’s a key part of modern evolutionary theory.
Certainly neutral drift is part of “modern evolutionary theory”; however, it has serious problems—which you seem to deny for whatever reason. But let’s move on.
Well, I'm not sure what those "serious problems" are. It is a fairly straightforward observation that the distribution of novel variants tends to peak near neutral, and that near-neutral variants have a sporting chance of replicating them selves in substantial numbers in the population, despite conferring little or no phenotypic advantage, and despite possibly being mildly deleterious, as I point out below:
What we now know is that the vast majority of new variants are near-neutral, which means that any given population at any given time represents a pool of phenotypic variants of which some variants may prove more successful than others in surviving when the enviroment changes.
This is probably true—given enough time for drift to work.
You seem to be stuck in your agency metaphor! "Drift" doesen't need "enough time" to "work" - it isn't an agent, any more than "positive selection" and "negative selection" are agents. They are simply ways of describing differential reproductive success. If a genetic variant makes no difference to the phenotype's chances of reproducing successfully, relative to that of its peers, we say that it is "neutral"; if does make a difference, we say that it is either "beneficial" or "deleterious". But these are not discrete categories, and all "drift" through the population or out of it. However, a variant the tends to enhance the probability of reproductive success will have a "bias" to its drift - will be a little more likely to become more prevalent than a neutral variant would be, and vice versa for a variant that tends to lead to reduced reproductive success. In other words the whole notion of "Natural selection" really needs to be seen as a stochastic "drift" process, in which some variants, in some environments, have greater or lesser chances of becoming more prevalent than others.
But, is the mere ‘presence’ of such ‘successful variants’ sufficient. Obviously the ‘presence’ of these putative ‘successful variants’ is necessary. But is it also sufficient for macroevolution to take place? We continue . . . .
This means that when we set up Darwinian computer models, we find, to our surprise, that complex functions evolve more readily than perhaps Darwin might have anticipated because necessary but neutral precursor variants are already likely to be there in the population.
Is this true? Do ‘Darwinian computer models’ actually show this? Behe and Snoke set up what they considered to be a ‘Darwinian computer model’ and they got no such result—in fact, it showed even NS, let alone neutral drift, was rather impotent to bring about meaningful changes.
Well, it won't always happen, clearly. The issue is whether it can happen. And Lenski et al, with AVIDA, showed that it can - i.e. that useful functions (useful to the organism) can evolve even if necessary intermediate steps are neutral or even deleterious. If there are no beneficial intermediate steps, then the function is clearly extremely unlikely to evolve, and Darwinian processes will work no better than a random draw. So the issue is not whether Darwinian process can work (we know they can) but whether they will work on the kind of fitness landscape that biological organisms inhabit. What AVIDA shows is that those fitness landscapes include landscapes with substantial plateaux and horizontal ridges, and even some ravines, not just landscapes in which all peaks are reachable by steady climb.
But let’s say that what you say is true. Well, what you’re saying is that the “necessary” neutral precursor (variant) is already likely there. Okay. But, as I said above, the variants are necessary, but are they sufficient to explain evolutionary change? This is where the whole idea of “fixation” comes in. You’re simply assuming that because this ‘beneficial’ (I know, ‘beneficial’ relative to the ‘current’ environment) is there, the rest automatically takes care of itself. But this is where, IMHO, you’re going wrong. The next step—which is ‘necessary’ and ‘sufficient’—is for this ‘beneficial variant’ to become fixed, wherein, every member of the population shares this variant.
No, "fixation" is not necessary. Indeed, the fixation of a neutral allele (and certainly of a deleterious one) is a problem for a population and thus for small populations where fixation is more likely. Fixation means that potentially useful variance has been lost. What makes it more likely that a useful function will evolve via neutral steps is simply that there are plenty of instances of those neutral steps, because the more opportunities there are for the next beneficial step to occur, the more likely it is to occur.
This is because, let’s say, a little kitten, born with a ‘variant’ that allows it to sniff out potential prey better than any of the other cats around, has a likely reproductive advantage over its litter mates as an adult. And so, too, would its offspring. But what if, one day, while still a kitten, it became distracted while crossing the street, and was run over. What, then, of this ‘beneficial variant’? What becomes of it? ANS: Nothing. It’s lost to the population.
Sure. That's what I mean by "selection" being a highly stochastic process. Many novel potentially beneficial variants may well be lost before their bearers even have a chance to reproduce. This is why it seems much more likely to me that most advantageous variants probably start as neutral variants, and become advantageous long after they have a history of substantial prevalence in the population. In other words that most "beneficial variants" are not novel at the time they become beneficial. I realise that this may be a someone unusual way of looking at Darwinian evolution, but it seems to me to make sense of what we know. More importantly, we observe it in the field (and lab) as what you call "microevolution" - the increased prevalence of already existing variants in response to environmental challenge. My point is that this feed of potentially useful (but currently neutral) variance is being constantly drip fed into the population and propagated by drift, so that "macroevolution" is simply the same process as "microevolution" but with a sustained movement in a single direction, rather than oscillation around a mean, and driven by sustained, rather than oscillating, environmental change.
This is a random event, and this randomness—that is, the potential for a single member of the population to be accidently lost, or, for whatever reason, not be able to furnish offspring; or for the offspring themselves to be killed off or reduced in number—all these random kind of events need to be overcome by the population. And that’s why population geneticists calculate such things as “fixation” times. Once a variant is incorporated into the entire population, it’s very likely not to be lost very easily. And now the population, as a whole, can work on acquiring—and fixing!—the next variant. But how long does this take? Well, Kimura did the calculations, and it turns out to be 1/4N_eff. And, hence, in 49.1 below, I include a sample calculation from Kimura’s magnus opus, “The Neutral Theory of Molecular Evolution”. It turns out that this is a very slow process. So……the mere appearance of a ‘beneficial variant’, though ‘necessary’, is not sufficient. Fixation is also a necessary, and somewhat sufficient condition for macroevolution to occur. The problem though is that this “fixation” process works way too slowly to account for the fossil record. This is only more so the problem when considering chimp—human differences (only 4 million years for all these “fixations” to occur!)
Well your error, IMO, is assuming that "fixation" is a necessary condition for selection. It isn't. Clearly, the more examplars of a novel sequence exist in a population (the more organisms that carry it) the more likely it is to hang around. This is what Kimura's equation tells us. So the earliest generations of a novel sequence are the most risky - the more established it becomes, the less likely it is to be lost, just as the more people are told a secret , the more difficult it is to suppress it. And the more examplars there are of a novel but neutral sequence, the more likely it is that a subsequent mutation that renders it beneficial is to happen simply because even rare events come close to inevitable given enough trials.
But what it does mean is that the old cliche that mutations are random and natural selection isn’t, is pretty well meaningless (and I wish people, especially Dawkins, would stop saying it!)
Dawkins is the leading apologist for the Darwinian cause. His entire argument falls apart if NS is considered to be ‘random’. Per your view, his arguments fall apart. So that leaves Darwinism with no leading apologist. So now what?
There are plenty of "apologist[s] for the Darwinian cause" but more to the point, it doesn't "need" apologists, it "needs" good scientists, which it has, aplenty, by which I mean there are plenty of evolutionary biologists in the world all doing sound work, refining and modifying evolutionary theory. However, Dawkins' argument doesn't "fall apart" of NS is considered "random". As I keep saying, "random" is a non-specific word that needs careful definition if it is to be used in a technical sense. I think Dawkins misuses it, in an attempt at simplification, and, in so doing, sets up a straw man. NS is not "random" in the sense that it refers to a process in which the survival of a given phenotype is not equiprobable for all genotypes. That's why it is called "selection". However, that doesn't mean that selection is not a stochastic process, as your kitten example shows. And "RM" is only "random" in the sense that novel variants are not more likely to be generated simply because they would be useful in the current environment. In that sense, novelty generation is "blind" to current needs. However, that does not mean all possible novel sequences are equiprobable (and this means even if we use the word "random" in the sense of "equiprobable" in relation to NS, it does not apply to RM). Novel sequences very similar to those of the parent sequences are much more probable than very different sequences, and as the parents are, by definition, fertile, the probability of producing fertile offspring is much more likely than the probability of producing non-fertile offspring. But far from "fall[ing] apart", correcting Dawkins' simplification means that the Darwinian proposal makes more sense, not less, and maps on to what we actually observe - novel variants that are strongly biased in favour of near-neutral phenotypic effects, and beneficial variants being pre-existing neutral variants now proving benefical in a novel environment, or possibly in a novel genoetype. I'll keep checking in here for replies to my posts, so if you have a reply, I'll gladly respond:)Elizabeth Liddle_{October 29, 2011
October
10
Oct
29
29
2011
09:11 AM
9
09
11
AM
PDT}

I'd better chip in here, as I have, in fact, desribed drift as a random walk :) But this is where we have to be so careful of these metaphors. The sense in which "drift" is a "random" walk is when it is the description of the changing prevalence, over time of a specific genotypic variant. However, the "drift" of a phenotype across a fitness landscape is not, as Dr Bot says, a random walk. I do think in these discussions it is very important to be clear when we are talking about genotypic variants (which, in a sexually reproducing population, can propagate through the population independently of other variants) and when we are talking about the fitness of a phenotype. So Dr Bot's example a "fitness landscape" defined as a two dimensional landscape along two genes. Let's call that "gene space". What he is showing is movement within that "gene space" by a phenotype, or, more strictly, by a population of phenotypes, the most prevalent pair of alleles within that population at any given time being given by the coordinates of X. Clearly, the traverse of the phenotype population across that "gene space" is not a "random walk" but is highly constrained by the topography of "gene space", and will go along in preference to down, and up in preference to along. However, if instead we examine the traverse of each allele (i.e. each gene "state" as defined as a point on the axes of "gene space") through "population space", defined with generations on one axis and prevalence on the the other axis, we will observe a "random walk", where the prevalence of that allele in the next generation may be equiprobable (if the allele confers no increase or decrease in probability of reproductive success relative to other alleles), biased in favor of greater prevalence (if it confers an increased probability of reproductive success) or biased in favour of reduced prevalence (if it confers a reduced probability of reproductive success). As Dr Bot says, you cannot carve neutrality of from selection. "Selection" is the name we give to the aspects of differential reproductive success that tends to keep populations "high" (on a ridge, up a slope) (I prefer myself to think of the landscape the other way up, with "gravity" as the analog of "selection", and populations tending to roll down slopes or along plateaux, but that's maybe just me :)), whereas "drift" can be used either to describe the "movement" of a population along a ridge or across a plateau, or to describe the propagation of alleles through a population. It is only in that last sense that it can be described as a "random walk",but usefully so, IMO, as it underlines the fact that what we call "selection" is a highly stochastic process, and one that is not simply a property of the allele (a simplifying assumption made in some population genetics model) but of the environment in which the phenotype has to survive, as well as the internal "environment" of the allele (other alleles with which it finds itself in combination in a given phenotype).Elizabeth Liddle_{October 29, 2011
October
10
Oct
29
29
2011
08:38 AM
8
08
38
AM
PDT}

DrBot: Please take your time :) I only hope to see your answer in the sea of unrelated posts this blog has become... In case, give me some (intelligent :) ) signal!gpuccio_{October 29, 2011
October
10
Oct
29
29
2011
07:52 AM
7
07
52
AM
PDT}

Thanks for the response, I'll try and respond myself but it may be a day or two before I can - I have family duties to attend to this weekend :)DrBot_{October 29, 2011
October
10
Oct
29
29
2011
07:04 AM
7
07
04
AM
PDT}

Prev 1 2 3 4 … 14 Next

You must be logged in to post a comment.

Leave a Reply