|
Author
|
Topic: Nature Refutes ID?: The Evolutionary Origin of Complex Features
|
charlie d.
Member
Member # 159
|
posted 21. May 2003 08:03
quote:
I also want to point out something about the detrimental mutations that were observed in the simulation. Notice that these fitness-reducing alterations were ALWAYS followed in the next generation by a fitness-enhancing change. Why is this? Why weren't the detrimental mutations allowed to sit awhile, maybe take over the population, before a beneficial mutation occurred?
I propose the following. The detrimental changes probably occurred in organisms that were fairly "fit" relative to their peers. Hence, they have some "capital" to play with and can afford to temporarily reduce their fitness. But, if they are unfit for long, they will be eliminated from the population. I'm sure many organisms had detrimental alterations which occurred and were eliminated. These organisms will never show up as winners at the end of the simulation. So this implies that once a detrimental mutation occurs, an organism's clock is ticking. It better find a way to get more fit--fast--so it can avoid elimination. In a practical sense, a mutation in the next generation is probably the only way to avoid elimination.
So yes--the organisms occasionally underwent a detrimental mutation. But sometimes they were able to "save" the situation by coming up with an even better, beneficial mutation immediately thereafter. But that's a far cry from an organism wallowing around in the depths of sub-optimality for generations and generations before stumbling on something that works. It's not surprising that we can look back over the evolutionary history and see organisms that made extremely temporary forays into areas of reduced functionality, and then bounced right back up the fitness peak. This doesn't really provide a general mechanism for crossing non-functional valleys, because the organisms that stay in these valleys will quickly be eliminated.
I fail to see how this is any different from real, biological scenarios. Organismal lineages suffer reductions in fitness all the time, in some cases they are lucky and recover through compensating mutations, but if these do not occur they become extinguished. Do you doubt that thousands upon thousands of lineages which suffered mutations within an AVIDA simulation became extinct, vs the few ones that survived and eventually recovered? If I may quote myself, I posted about this before: quote:
downward fitness variation is of course very common in biological organisms, and even a 50% decrease in fitness does not mean immediate extinction (that would only occur for fitness 0). According to population genetics, the average time (generations) to fixation of a new allele is 2/s*ln(2N), with s the selection coefficient (in this case -0.5) and N the population size. I have not thought through whether this can be applied straight to the AVIDA case, but if it can, a genotype with fitness 0.5 (s=-0.5) in a population with N=3600 would go extinct in 36 generations: quite some time for new mutations to occur. Even more so if the 50% decrease occurs in a low frequency genotype with already high fitness - in that case it may still result in an above-average fitness (>1), and the mutant genotype would actually keep expanding in the population (until the average fitness surpasses it because of the expansion in the higher fitness genotype, that is). Everything is relative.
In this respect, AVIDA is not a bad model of natural situations, as far as I can tell.
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 21. May 2003 08:52
RBH,
Quote: Once again warren makes the unsubstantiated claim that there are biological processes and structures that are supposed by biologists to have occurred by means of the mechanisms of evolution, but that developed at too high a rate for those mechanisms to plausibly account for them.
While I support the claim that evolutionary change in biological systems is far to fast and powerful to be explained by the very slow and very inefficient random mutation and natural selection processes, that was not the focus of my comments yesterday. My comments were focused more on pointing out the shortcomings of the study being discussed.
The Lenski study is presented as an ‘it could have happened that way’ demonstration that ‘mutation and natural selection processes could, in a particular type of abstract mathematical universe, produce some ‘evolution of complexity’. No where, as far as I can tell, did they attempt to define or quantify how much complexity their system created nor did they attempt to quantify how much ‘mutation-selection’ processing power was required to produce the undefined increase in complexity. Finally, the study failed to provide a precise definition of a ‘random mutation-natural selection’ search routine.
Because the Lenski study fails to define and quantify any of the key concepts, it is just as legitimate to interpret the study as showing ‘natural selection and random mutation is too inefficient a process to explain the evolution of complexity’ as it is to conclude ‘natural selection and random mutation is sufficiently powerful to explain the evolution of complexity’. The Lenski study demonstrates that if you run a certain type of program, you can generate a certain type of results. Because the study fails to define and quantify key variables, it is not possible to draw any meaningful conclusion about evolution of complexity in an abstract mathematical universe. If you can not draw conclusions regarding evolution in an artificial environment, you certainly can not draw any conclusions regarding evolution in biological systems.
John Bracht,
I agree with your observation that there appears to be a double standard with respect to ‘biological realism’. It is interesting to note that the use of double standard can create the erroneous impression that ‘the evidence supports Darwinian theories’. If one standard of biological realism is used to evaluate pro-Darwinian analysis, such as the Lenski study, then these articles will get published. If another standard is used to evaluate analysis suggesting anti-Darwinian conclusions, then these articles are unlikely to get published. The impression created is that essentially all analysis passing peer review supports Darwinian theories.
IP: Logged
|
|
YZ2
Member
Member # 91
|
posted 21. May 2003 10:27
I think Argon has raised the more important question.
Argon:
--------------------------------------------------------------------------------
I think we all agree with the truism that evolutionary mechanisms can only create evolvable functions, regardless of whether we subjectively consider the routes "smooth" or "bumpy". Withholding judgement about the relative poopiness of any particular simulation for now, can we address the question of whether the derived EQU functions are IC or not? Are they specified or not? A significant question is whether evolutionary mechanisms can generate IC systems (apparently yes), or specified complexity (dependent on the pathway, apparently).
--------------------------------------------------------------------------------
In reply to this:
I personally think that EQU is IC and ID, and that it is generated from an evolutionary process at least in this case (efficent or not), and there is no need for inconsistency and conflicts. Of course there are more people here more experienced than I am that can further determine these claims. My reasoning is this.
If we accept an ID theory with a possible evolutionary mechanism, then it is at least as consistent and fruitful as a designer-free evolutionary theory, since the former includes the latter. In this framework, there is no apparent scientific reason for conflict between the two. They address different aspects of the empirical phenomenon, overlapping that it may be, and can pose different scientific questions. The ID approach focuses more on the design aspect of a phenomenon (the selection function side if you will), and the evolutionary approach focuses more on the process aspect (the fitness function side). So it is unfortunate for polarized positions that can hinder the advancement of science. Now it is not entirely accurate to say that evolution is both a theory and a fact. A theory is a collection of consistent assertions of possibility, supported by the available evidence. The asserted possibilities give a theory its predictive power. On the other hand, an asserted event in the past that is consistent with a theory does not necessarily mean that it has actually occurred. Whether it has occurred is either true or false, but not may be possible. In establishing whether complexity in life has evolved in the past cannot be convincingly determined by just a theory. It can only be determined by a legal argument, based on the evidence. Its doubt or belief can then be evaluated.
[ 21. May 2003, 11:58: Message edited by: YZ2 ]
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 21. May 2003 15:03
Micah gave us some hint of how we may go abouth determining if the systems is IC
quote:
Behe's definition of Irreducible Complexity:
A single system composed of several well-matched, interacting parts that contribute to the basic function of the system, wherein the removal of any one of the parts causes the system to effectively cease functioning.
Under this definition EQU would seem to be IC
Although Micah argues that "To sum up: The EQU function does not meet even Behe's original defintion!!! When you remove one NAND primitive, you are left with four others that perform some other logic operation and which has a functional value of 16 (as opposed to 32 in the EQU)."
So Micah suggests that original function may not be important. But as Dembski argued when trying to rectify some problems in Behe's claims:
quote:
A system is irreducibly complex in Behe's sense if all its parts are indispensable to preserving the system's basic function. That an irreducibly complex system may have subsystems that have functions of their own (functions distinct from that of the original system) is therefore allowed in the definition.
I would say that EQU is likely to be IC. But it would be helpful if we were to get some feedback from Dembski or Behe on this. Micah, do you have access to these people? After all, science can only progress when all agree on the definition of the terms and IC so far seems somewhat hard to pin down. Even in our discussions IC seems to have evolved to "that which cannot evolve"
IP: Logged
|
|
yersinia
Member
Member # 324
|
posted 21. May 2003 17:50
Hey John,
I think the reason that they used EQU etc. as the complex functions for which rewards in terms of "food" (SIPs) were given include:
1) It is easy to test whether or not the EQU function has been achieved, and such a test is independent of analysis of the function (unlike, say "METHINKS..." which requires the algorithm to read the string and compare to the pre-specified string goal).
2) Doing what you seem to be asking for (a complex function advantageous in real life) would require a simulation of a substantial chunk of the environment. E.g., to simulation the evolution of a flagellum, you would have to simulate fluid dynamics, ecological interactions with predators and prey, spatially and temporally varying nutrient gradients, etc., as well as similar interactions relevant for transport systems etc. You could spend your whole life trying to model a single bacterial population in such a fashion. Each generation might take years to compute if it could be done at all. The number of "calculations" that occur in a real ecological history is immense.
There may be various ways to simplify the above but it is far from trivial.
So, in order to answer a specific question, the authors assumed two things which are known to be true in biology:
a) Complex systems have net reproductive benefits
b) Subsets of these systems, with different functions, also have net reproductive benefits
...and asked, given (a) and (b), can complex multiple-parts-required systems evolve?
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 21. May 2003 22:24
Pim quotes Micah as saying quote:
Although Micah argues that "To sum up: The EQU function does not meet even Behe's original defintion!!! When you remove one NAND primitive, you are left with four others that perform some other logic operation and which has a functional value of 16 (as opposed to 32 in the EQU)."
I'm going to inject some pedantry here. There's growing fuzz surrounding what it is that was evolved, and what it is that is irreducibly complex by Behe's operational definition, the knockout procedure. As far as I know, that's not yet been abandoned, has it?
What evolved were strings of assembly language instructions that perform EQU. EQU itself didn't evolve; genotypes (instruction sequences) capable of performing EQU evolved, just as in the flagellum what evolved were strings of genes that code for the protein assemblage that provides motility.
Now, what is it that is "irreducibly complex"? Is it the function performed by a structure, or is it the phenotypic structure, or is it the gene string that codes for the structure? For the flagellum, "specification" is at a fairly high level - rotors, stators, cilia. The flagellum is also IC at that level: Knock any one of them out and the flagellum no longer functions as a flagellum. But the calculation of probabilities is performed (by Dembski, at least) at the protein level - the flagellum displays specified complexity because the probability of chance assembly of proteins in that particular configuration is too low to have plausibly occurred. Thus one is to infer it was designed. Dembski's analysis does not consider the genetic instructions that code for the protein structure of the flagellum; they are irrelevant in his analysis.
In the case of the Avida simulation, we have just two levels rather than three: coded instructions and phenotypic functions. (Well, in the computer there's actually a third level lurking out of sight, the machine language level.) The phenotypic function, "EQU," is irreducibly complex because if one knocks out any of the terms or operators of [(A ^ B) or (~A ^ ~B)], the minimal formal expression of EQU, it is no longer EQU.
What actually perform logic functions in Avida are strings of assembly language instructions, not merely several NANDs. By Behe's operational definition of IC, the 23 different instruction strings that evolved to perform EQU were also irreducibly complex. In every case the knockout procedure applied instruction by instruction destroyed the function of performing EQU. Thus at both levels, the formal symbolic expression of EQU and the instruction sequences that performed the function EQU, the system is irreducibly complex.
At which level does one calculate the probability necessary to ascertain whether the simulation produced "specified complexity"? "Specified" is no problem - the EQU function is clearly independently specifiable. But if the question is about what actually evolved, it's the sequences of instructions that perform EQU, and that seems like the appropriate level at which John should calculate his allegedly high probability of chance assembly by shuffling around assembly language instructions. As I've said before, that probability calculation must include the fact that 23 different irreducibly complex instruction strings evolved to perform EQU in 50 attempts.
John wrote quote:
I also want to point out something about the detrimental mutations that were observed in the simulation. Notice that these fitness-reducing alterations were ALWAYS followed in the next generation by a fitness-enhancing change. Why is this? Why weren't the detrimental mutations allowed to sit awhile, maybe take over the population, before a beneficial mutation occurred?
But they were allowed to "sit awhile." They could sit as long as they were just reproductively successful enough to avoid extinction. They didn't take over the population because they were up against more fit cousins. In Avida there are typically a fair-sized number, sometimes a large number, of different "species" simultaneously present and competing, and they vary in fitness. Further, fitness-reducing alterations were not always followed "in the next generation" by a fitness enhancing change. There are two aspects of the misunderstanding that leads to that error. John does not understand the "step" metric. It is not a measure of generations. Second, in the lineages in which sequences capable of performing EQU evolved, a deleterious mutation was frequently, but not ALWAYS, followed after some time by an advantageous mutation that was the next "step", but that was not the case in all lineages. You really need to read that part (page 141-2) carefully, John.
RBH
[ 21. May 2003, 23:13: Message edited by: RBH ]
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 22. May 2003 07:43
I'm not sure that this thread is going anywhere at this point, and probably won't be contributing beyond this post, but I just wanted to throw in this challenge to the mix (not my idea).
Could the Avida system evolve a simple multiplication function that could multiply any two 32 bit strings together. After all, there are millions of these operations being performed in your CPU each time you make a post to this thread.
A few other questions:
1. If EQU is IC, then could someone be explicit about what the parts of the EQU are? Are we equating an entire self-replicating Avida program with EQU? Or is it a subsystem that is readily identifiable and separable into Behe's "several well-matched" parts?
2.If EQU is IC (which I don't go for), then why the 50% success rate and why no contingency in what can be achieved? Why no other 2-input logic functions?
I think this scould be a good example of Mike Gene's (and others) notion of constrained, teleological evolution. In fact, if evolutionary engineering ever becomes successful, it will be a form of highly constrained evolution, which in my mind is an exciting engineering task, but certainly not a glimpse into the workings of real-world biological dynamics.
One final question, has anyone been able to find the "better than human" organism that solved EQU in 17 necessary Avida instructions?
[ 22. May 2003, 07:44: Message edited by: Micah Sparacio ]
IP: Logged
|
|
Argon
Member
Member # 276
|
posted 22. May 2003 11:48
Micah, regarding your most recent questions:
Part of the confusion results from the fact that even Behe himself uses several different definitions for "IC" and doesn't always distinguish between them. But if one begins with Behe's DBB, and uses with the original, operational criteria presented there, I think the answers are relatively straightforward in this instance.
But the very second one tries to mix-'n-match definitions, one goes down in flames because they are not interchangeable.
[ 22. May 2003, 11:51: Message edited by: Argon ]
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 22. May 2003 12:34
This is the question I'm most interested in.
1. If EQU is IC, then could someone be explicit about what the "several well matched parts" of the EQU are? Are we equating an entire self-replicating Avida program with EQU? Or is it a subsystem that is readily identifiable and separable into Behe's "several well-matched" parts?
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 22. May 2003 14:36
Micah,
As I read the paper (and I just reread it once again), what was tested in the knockout procedure, which is Behe's operational definition of irreducible complexity, were the individual instructions in the assembly language programs that evolved to perform EQU in Avida: quote:
...we systematically replaced each existing instruction with a null instruction in the first genotype able to perform EQU. We scored each null mutant for which functions were lost and which remained, and the data from all the [null] mutants were combined to produce an array showing the relationship between genome sequences and phenotypic properties (Fig. 4). (p. 141)
I interpret this to mean that they were not unconcerned in this analysis with the replication function, but were primarily interested in the logic functions. Nevertheless, they checked to see if replication was lost due to the null replacements as well.
Figure 4 shows that of the 60 total instructions in the 'case study' genome, 8 were critical to replication - that function was lost if any one of those instructions was replaced by a null instruction. Of course, replication was a pre-existing capability, so that particular finding tells us little.
Of the 52 remaining instructions, 12 had no effect on the performance of any function when replaced by a null instruction. They might be (cautiously) interpreted as "junk" instructions, though that might be a stretch.
Finally, of the 48 'non-junk' instructions, 35 were part of what one must call the "irreducible core" of the program's functioning to perform EQU, in that replacing any one of them with a null instruction resulted in the loss of the ability to perform EQU. Three of those were also involved in replication, but performing EQU does not require replication. That suggests that the capability to perfom EQU 'borrowed' some of the same instructions involved in replication - those instructions apparently performed a dual role, perhaps by producing side effects (a value left in a register, for example) that the other instructions - those not involved in replication - utilized in performing EQU.
Those 35 instructions then constitute the "several well matched parts" that comprise the irreducible core of the program that evolved to perform EQU as determined by Behe's operational definition of "irreducible complexity," loss of the function if a part is knocked out.
One of the frustrating things about the several definitions of IC that are floating around is that they lack operational interpretations - how the heck does one apply them to arrive at an unequivocal researcher-independent classification of a phenomenon as IC or not-IC? The only operational definition that has been offered is Behe's knockout procedure, and so that is the only definition that can be used in actual scientific research, as distinguished from philosophical blathering. I have no quarrel with philosophers - indeed, I had a friendly conversation with one just 45 minutes ago in front of the post office. But if they are going to offer definitions by means of which scientists are to measure and classify phenomena, they have to pay some attention to the operational (researcher-independent measurement and/or classification methods) requirements that science requires of those definitions.
The knockout procedure as a classification criterion for IC vs. non-IC is an appropriate example. Anyone can do it to any phenomenon and arrive at pretty consistent classifications across people; whether a given phenomenon is classified as IC or non-IC does not depend heaviily on the particular characteristics, beliefs, or predispositions of the person doing the classifying. (There is still some fuzz surrounding the knockout procedure, though. For example, at what level of analysis does one apply the procedure? If one adopts the level of atoms, for example, Behe's mousetrap is not IC in that one could knock out any one atom of a mousetrap with no perceptible effect on its function. Behe has offered no principled way of determining the level of application of his operational procedure for determining ICness.)
The "well matched" phrase is not a good example of an operational classification rule because there is no hint of how to assess or measure the property 'well matchedness' and thus no way to consistently classify phenomena in a researcher-independent way. While I don't want to regress all the way back to P. W. Bridgeman's (1927) radical operationalism, nevertheless one has to have agreed measurement procedures that allow classifications that don't depend on which researcher is doing the classifying, and "well matched" doesn't do that.
RBH
OK, I'm done editing now: it's safe to respond! ![[Smile]](smile.gif)
[ 22. May 2003, 15:08: Message edited by: RBH ]
IP: Logged
|
|
Argon
Member
Member # 276
|
posted 22. May 2003 18:20
RBH writes:
quote:
The "well matched" phrase is not a good example of an operational classification rule because there is no hint of how to assess or measure the property 'well matchedness' and thus no way to consistently classify phenomena in a researcher-independent way.
I concur. "Function" is also a description that could present a problem in the IC definition.
All that aside, I think an "IC case" could be made for the two-step process by which streptomycin resistance is stably acquired in populations of E. coli and Salmonalla typhiumurium. In this case a mutation in rpsL (a ribosomal protein), which renders the bacterium resistant to the streptomycin, is selected during exposure to the antibiotic. Some of these resistance mutations carry a cost that reduces the bacterium's ability to compete with the normal, "wild-type" strains when grown in the absence of the antibiotic. In some cases the cells simply grow slower while in others, their pathogenic virulence is lost. Yet when resistant strains are propagated for hundreds of generations in environments free from streptomycin, many do not revert to the drug-sensitive state. Instead, we find these strains acquire secondary, compensatory mutations which ameliorate some of the problems produced with original mutation in rpsL. I don't think it is too much of a stretch to consider the interactions between ribosomal proteins as being "well-matched".
Interestingly, when strains were constructed that carried only the compensatory mutations (no mutant, strep-resistant rpsL), researchers observed defects in growth rates or virulence. Only when combined do the two mutations compensate and confer resistance.
Could this resistance/compensatory system be considered IC? Possibly. Consider what happens when either of the matched mutations are corrected to the original sequences, thereby "knocking-out" the function conferred by that mutation. Loss of either mutation has a negative effect on the ribosomal activity. Losing the rpsL mutations also results in a loss of streptomycin resistance. And notice how this case parallels the EQU simulation we've been discussing. These mutations are not acquired in a single step -- Selection for growth in the presence of streptomycin must happen first, followed by competition against either wild-type strains (when grown in the absence of the antibiotic) or against "fellow" strains that exist when streptomycin is present. The odds of specifically mutating both genes simultaneously are very slim (minimally, two specific point mutations in different genes). By themselves, these individual mutations are unlikely to persist in the environment at significant levels. And yet, this system of compensatory mutations has been demonstrated repeatedly in experiments.
Here are some useful references:
The biological cost of streptomycin resistance - DI Andersson & BR Levin
http://www.emory.edu/BIOLOGY/EcLF/pubs/99costres.pdf
Compensatory mutations, antibiotic resistance and the population genetics of adaptive evolution in bacteria - BR Levin, V Perrot & N Walker
http://www.emory.edu/BIOLOGY/EcLF/pubs/00compmut.pdf
BTW - Richard Lenski's name occasionally pops up in these references.
IP: Logged
|
|
Roger R
Member
Member # 200
|
posted 22. May 2003 20:41
RBH writes:
quote:
The knockout procedure as a classification criterion for IC vs. non-IC is an appropriate example. . . .
The "well matched" phrase is not a good example of an operational classification rule because there is no hint of how to assess or measure the property 'well matchedness' and thus no way to consistently classify phenomena in a researcher-independent way. . . .
The problem with your dependence on the knockout criteria is that it alone is not sufficient to define IC. That "well matched" is open to some interpretation is true enough, but it is also true that you can't just ignore it and claim "knockout" is the litmus test for IC. "Well matched" is critical to understanding why EQU is not IC, and not analogous to the bacterial flagellum.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 22. May 2003 21:12
Roger, feel free to revise the definition of IC but if knockout is not the lithmus test but well-matched then I wonder if much of the original argument of Behe's remains. And in fact the EQU seems to have several well matched systems without any one of such a system EQU would stop functioning.
May I invite Roger to present us his argument as to how "well matched" eliminates EQU as being IC?
I find it somewhat frustrating how flexible IC is to avoid any objective definition. Surely any claim that IC systems defy evolution seem to become harder to maintain when IC itself evolves so quickly ![[Wink]](wink.gif)
Speaking of Lenski and compensatory mutations, here a Recent paper by Wilke, lenski and Adami
[ 22. May 2003, 21:22: Message edited by: Pim van Meurs ]
IP: Logged
|
|
Roger R
Member
Member # 200
|
posted 22. May 2003 21:27
I've revised NOTHING. The well-matched is part of the original Behe definition. It isn't a matter of knockout vs well-matched. Both are important parts of the concept Behe is relating, and you can't just drop part of the definition because it is inconvenient.
Again, the goalposts are being moved by the same crowd decrying such actions.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 22. May 2003 22:14
Roger wrote quote:
The problem with your dependence on the knockout criteria is that it alone is not sufficient to define IC. That "well matched" is open to some interpretation is true enough, but it is also true that you can't just ignore it and claim "knockout" is the litmus test for IC. "Well matched" is critical to understanding why EQU is not IC, and not analogous to the bacterial flagellum.
But it is not my dependence on the knockout criterion, it's Behe's. It is his operational definition, the only objective way one can operationally classify a system as IC or not. You see, Roger, in order to meet the requirement for intersubjective replicability in science, which requires researcher-independent measurement methods, there has to be a way to objectively measure or classify objects. Looking at a system and saying, "Gee, those parts look well-matched to me" is not an objective measure of 'well matchedness.' The only classification operation that Behe (or anyone else, for that matter) has offered to discriminate IC from non-IC systems is the knockout operation. That was actually sort of encouraging - it meant that Behe had not altogether lost his bearings as a scientist. But if his defenders disavow the only operational definition of ICness that has been offered then the term is wholly useless from the point of view of doing research on IC systems, because there is no way to tell whether a system is IC or not except by subjective impressions. And that ain't science, Roger. I encourage you to read up on the rise and fall of N-rays for an example of what happens to such notions.
RBH
IP: Logged
|
|
|
Printer-friendly view of this topic
