|
Author
|
Topic: Royal Truman: Avida, a biologically unrealistic model for neo-Darwinian Theory
|
RBH
Member
Member # 380
|
posted 30. January 2004 21:58
I'm not at all sure this is a sidetrack. It's at least peripherally relevant.
charlie d wrote quote:
I don't want to drag the discussion off topic, but I would like to comment on RBH's statement that AVIDA does not simulate anything. In fact, I think AVIDA is indeed a simulation, simplified as it is, of a population of imperfect replicators under selection. That this corresponds to certain salient features of biological organisms (or better, their genetic components), which are considered necessary for biological evolution by darwinian mechanims, is what makes AVIDA interesting after all.
The designers of Avida pretty clearly see Avida as an alternative venue in which evolution occurs. They built it to instantiate (some of) the mechanisms thought to universally characterize systems capable of evolving - your "population of imperfect replicators under selection" - but they did not set out to simulate any specific organic biological system or even a generic organic biological system. Rather they are providing a platform in which real evolution actually occurs in order to do comparative research. The digital critters in Avida are "imperfect replicators under selection" just as much as a bunch of E. coli in a Petri dish might be imperfect replicators under selection!
That at least some properties and dynamics of a population of digital critters evolving in Avida correspond to some properties and dynamics of a population of organic creatures evolving in the physical world is a statement about the generality of evolutionary processes and mechanisms, not a statement about the fidelity of a "simulation." Evolution in populations of Avidian critters is no more a simulation of organic evolution than evolution in populations of organic creatures is a simulation of Avidian evolution.
This is a very loose analogy, but having Avida is more akin to finding the fossil of the wing of a pretty weird and primitive creature to include in a comparative morphological study of fossil forelimbs used in flight, than it is a "simulation" of organic wings. In my view, that is what makes it interesting. Heretofore we had only one flying system; now we have two and can therefore do comparative research on evolution in populations of Avidian and organic critters.
RBH
Added in edit: I mentioned in an earlier posting that I started an Avida run using the configuration files of an old run in which mutations rates were set very low and Avida's analogue of lateral gene transfer was enabled. I restarted that run again a bit later, setting the run to terminate at 500,000 updates rather than the default 50,000. It's still grinding away (I said it's an old slow machine!), and at 275,000 updates it has lineages that among them perform input-output operations that correspond to all 9 of the logic functions. The average critter's program length is 197 instructions; the average number of executed instructions per critter is 106. So the critters have acquired an average of 41 "junk" instructions more than the original 50 nop-B junk instructions in the Ancestor critter and have acquired an average of 88 additional executed instructions over the 18 replication instructions the Ancestor started with. There is still considerable diversity of instruction strings (assembly language programs) in the population, with no single string coming even close to dominating the population. I'm not saving out the individual lineages - that machine has way too little free disk space - but I'm saving summary data on a couple of dozen variables every 50 updates and I'm looking forward to seeing what that looks like after 500,000 updates. But it's already obvious that even with mutations rates one-tenth those used in the Lenski, et al, study, and with the analogue of lateral gene transfer enabled, those pesky irreducibly complex critters still evolve.
[ 30. January 2004, 22:32: Message edited by: RBH ]
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 30. January 2004 23:15
RBH: But it's already obvious that even with mutations rates one-tenth those used in the Lenski, et al, study, and with the analogue of lateral gene transfer enabled, those pesky irreducibly complex critters still evolve.
Fascinating findings, although not really that unexpected. The paper by Lenski et al has contributed significantly to our knowledge of how natural processes can create 'IC' or 'CSI'.
I wonder how MESA is doing?
[ 30. January 2004, 23:27: Message edited by: Pim van Meurs ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 31. January 2004 08:54
RBH: I see your point, but the salient features of AVIDA were chosen a priori to reproduce those of the only other evolving system we know about, i.e. biological organisms. Indeed, they were chosen to duplicate the minimal features required for evolution, as we can infer them from natural evolving systems on Earth (which raises an interesting question: can there be other kinds of evolving system, based on different principles?). The fact that crucial aspects of AVIDA can parallel equivalent aspects of biology (so that we can fairly give them the same names: mutation, replication, "junk instruction", "lateral information transfer" etc) is a testament of this. Whether that merits the term "simulation" (intended as, per the AHDEL, "representation of the operation or features of one process or system through the use of another") is a matter of definitions.
[ 31. January 2004, 08:55: Message edited by: charlie d. ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 31. January 2004 15:11
charlie: OK, I see your point, too. I don't think we disagree. The Petri dish came first. ![[Smile]](smile.gif)
RBH
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 08. February 2004 16:46
Thanks to Rex Kerr for some constructive comments.
The issue the authors seek to solve is surmounting prohibitive statistical jumps needed if neo-Darwinian processes are to generate novel cellular functions. The key assumption is the existence of a ladder of convenient 'rungs' which are statistically easy to reach. I point out we are not interested in an example of such a concept 'in principle': the notion itself is easy enough to understand and hardly new. The question is biological relevance: IS the random genetic (or protein) sequence space loaded with such useful functions (‘rungs’), and are they easy to reach by the process of random mutations and natural selection. I deny this.
[Rex]: “At some point, perhaps only very small genomes existed--but is Avida a model of this?”
Yes, of course. The authors work with a 50 instruction genome supposedly able to perform autonomously all metabolic and reproductive needs. Their example represents a total genome less than one sixth the size of an average length gene. Worse still, most of this genome is assumed to be superfluous, and thus available of evolutionary experiments.
[Rex]: “That's not what the article was about. Avida isn't intended to simulate abiogenesis,…”
Of course not! No one, especially not me, is making any such claim!! Where did this statement come from?!
[Rex]: “and the authors make no claims that it is, so criticizing it for not doing this is silly.”
Nobody is making any such criticism. Abiogenesis faces a whole flood of problems unrelated to our topic.
[Rex]: “(Not to mention that the criticism seems to assume without argument that "very small genomes" must have been very fragile to mutation; I would think they'd be robust to mutation but inefficient and fragile to environmental variation.)”
My arguments should have been easy to follow. I pointed out that the real reproductive work is being carried out not by the organisms but the computer hardware plus program plus electrical current provided by the authors, and these are not subject to mutational damage.
Secondly, the small organisms would quickly eliminate all genetic material not immediately needed to survive. Thereafter, a high proportion of mutations on the the streamlined genomes would be deleterious.
[Rex]: “They colonize rapidly, but can be supplanted by a larger number of tougher and/or more efficient slower-growing organisms. (Take trees vs. grass, for example.) So while I agree that the unchecked exponential growth model is problematic for evolution of complex functions in the manner that Avida does it…”
The more complex and slower reproducing organisms are compensated for by additional useful functions which enhance their survival. In the case of Avida organisms, I pointed out that the organisms would have exactly the same functions initially. Those lineages with less superfluous genetic material would have be selectively favored and would quickly outreproduce the other lineages. The tiny Avida organisms, like small bacteria, don’t posses things such as a complex immune system like we do.
[Rex]:”Note: I don't think it's correct that genome enlargement is rewarded; it simply isn't penalized. If you have evidence to the contrary, please supply the evidence in place of the claim.”
From the paper, p. 140: ‘First, each organism receives SNIPs in proportion to its genome length.’
This is independent of evolved logic functions. The above sentence then continues: ‘Second, an organism can obtain further SIPs by performing one- and two-input logic operations on 32-bit strings’
What Rex needs to understand, is that the authors have chosen the kinds of organisms most likely able to produce novel cellular functions via neo-Darwinian processes: very small, rapidly reproducing creatures, possessing lots of junk DNA, able to survive very high mutation rates. I am merely accepting their chosen scenario and pointing out why this does not work once we insist on a minimum of biological realism.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 08. February 2004 18:15
Truman: I point out we are not interested in an example of such a concept 'in principle': the notion itself is easy enough to understand and hardly new. The question is biological relevance: IS the random genetic (or protein) sequence space loaded with such useful functions (‘rungs’), and are they easy to reach by the process of random mutations and natural selection. I deny this.
This may be Truman's perspective but the more interesting goal is to show how mutation and selection can in fact lead to increase in information and complexity and in fact can create IC/CSI. Whether or not these examples are biologically relevant is of lesser concern than addressing claims based on elimination which are based on the assertion that natural processes (regularity and chance) cannot create CSI. If it can be shown that contrary to this assertion natural processes can generate CSI, can increase information then any inferences based on elimination are doomed to failure in principle.
Truman: Thereafter, a high proportion of mutations on the the streamlined genomes would be deleterious.
Actually the highest proportion would be neutral. In fact the combination of neutrality and random mutations/selection make for a powerful combination as neutrality increases both robustness as well as evolvability.
Truman: What Rex needs to understand, is that the authors have chosen the kinds of organisms most likely able to produce novel cellular functions via neo-Darwinian processes: very small, rapidly reproducing creatures, possessing lots of junk DNA, able to survive very high mutation rates. I am merely accepting their chosen scenario and pointing out why this does not work once we insist on a minimum of biological realism.
So in short, Truman's article does not really address the stated purposes and goals of the researchers but rather argue that these approaches do not work for simulations more relevant to biology. In fact I would argue that the latter one is stated more as a premise than as a conclusion, certainly it cannot be based on Avida perse. In other words, the idea that RMNS cannot work on biologically relevant systems to increase information and complexity remains so far unexplored in much detail and contrary to what the (indirect) evidence suggests.
Tom Schneider responds
Truman: The question is biological relevance: IS the random genetic (or protein) sequence space loaded with such useful functions (‘rungs’), and are they easy to reach by the process of random mutations and natural selection. I deny this.
And yet the data seem to support such. Check out the work of Schuster, Stadler, Fontana and others on the neutrality of RNA and Protein space. Their findings contradict your denial I would argue.
Let me provide you with some references to start from:
Shaping Space: The Possible and the Attainable in RNA Genotype-Phenotype Mapping (1998) Fontana and Schuster J theor Biol 194 491-515
Molecular Insights into Evolution of Phenotypes (2000) Schuster Evolutionary Dynamics — Exploring the Interplay of Accident, Selection, Neutrality, and Function Edited by J. P. Crutchfield and P. Schuster, Oxford Univ. Press
Neutrality: A Necessity for Self-Adaptation. Proceedings of the Congress on Evolutionary Computation (CEC 2002), 1354-1359 Toussaint and Igel
[ 08. February 2004, 18:42: Message edited by: Pim van Meurs ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 08. February 2004 21:11
There is an aspect of chemistry and biochemistry that is prone to lead intuitions astray when it comes to life and evolution. When building macroscale machines out of wood, metal, etc., one finds that parts are not cross-reactive. A screw will go into a screwhole made for it, and really won't do anything else, for instance. But biochemistry is not like this; parts are terribly cross-reactive. Any time you use an affinity column or run a two-hybrid experiment, you get proteins sticking together that you never expected (or wanted, or are relevant to biological function as far as we know). Heat shock proteins aid survival by, among other things, binding to misfolded proteins not because misfolded proteins are useless but because they are too sticky; they react with too much. People have done screens to generate micromolar-affinity binding peptides and found that you only have to screen millions of largely random peptides in order to find something of that affinity. Many common drugs were found precisely because small chemicals interact with so much--and this is also the cause of most of their side-effects, as they interact with multiple systems in the body. Because of this, intuitions about throwing stuff together in a machine shop--or even in a conventional computer program--are in enormous danger of failing when it comes to biochemistry. I know enough to know that my intuition is a very poor guide at best; if there is no experimental data, the question is better considered unanswered.
I think this faulty intuition is behind a large part of the claims made by ID proponents--even the biochemists who should know better! It is understandable, though: if you look at a system that is tightly specified through millions of years of refinement, it is sometimes easy to forget how cross-reactive the starting materials are, in general.
Now, for specific comments.
I don't see that Lenski et al. really addressed the relationship between the size of the jumps needed in their system and the size needed in biological systems. That is still an open question, and they explicitly state what they have shown and acknowledge the difference:
quote:
Our experiments demonstrate the validity of the hypothesis, first articulated by Darwin and supported today by comparative and experimental evidence, that complex features generally
evolve by modifying existing structures and functions . . . Of course, digital organisms differ from organic life in their genetic constitution, metabolic activities and physical environments. [Nature 423:143]
They aren't addressing whether the convenient rungs exist in biology--they are demonstrating that if you have rungs, they can be climbed. ID proponents might take this as a result in their favor, in fact: see, you need rungs!
As to whether Avida is a model of an organism with a very small genome, I don't think "Yes, of course" is the right answer. Avida has a very small digital genome. However, it does not, as you point out, have to support the variety of critical functions that a presumed small-genome life-form would have to support. But this is, again, not what the research is about. There are very, very good reasons to simplify the problem--for instance, that your simulation will take much too long to run if you don't. Avida was therefore designed to test evolution of complex operations from simple ones and not to be a faithful model of a small-genome living organism. Note that the paper never makes claims about the size of genome in a biological organism that Avida might correspond to.
And if Truman doesn't think Avida has anything to do with abiogenesis, I wonder why he thinks that genomes should be small. Selection for smallness, perhaps?
Truman hasn't provided evidence that the extreme selection he envisions for highly streamlined genomes is biologically realistic, or that there are biological systems under these kind of pressures that nonetheless are thought to have developed complex systems, or that selective effects overcome entropic effects. Indeed, his assumptions would predict that Mycoplasma is, at this very moment, dying out due to an already minimalized genome and unavoidable accrual of mutations. Funny thing is, it seems to be doing just fine, which is an indication that some of the assumptions used are wrong.
quote:
[Rex]:”Note: I don't think it's correct that genome enlargement is rewarded; it simply isn't penalized. If you have evidence to the contrary, please supply the evidence in place of the claim.”
[Truman]:From the paper, p. 140: ‘First, each organism receives SNIPs in proportion to its genome length.’
Very good. Now, how many SIPs does it take to copy an organism? How does this vary, if at all, with genome length?
quote:
What Rex needs to understand, is that the authors have chosen the kinds of organisms most likely able to produce novel cellular functions via neo-Darwinian processes: very small, rapidly reproducing creatures, possessing lots of junk DNA, able to survive very high mutation rates.
Indeed they have, as we don't have infinite computational power, and we want to start by establishing the basic concept that complex functions can be built from simple ones given a variety of indirect pathways.
Since they didn't claim the method was highly biologically realistic, pointing out that it is not highly biologically realistic isn't a very interesting criticism. The interesting criticism (if supportable) would be: their assumptions are so generous--and the development of complex functions requires such generosity in assumptions--that biological systems cannot generate complex functions in anything resembling an Avida-like way.
So far, the arguments to that effect seem highly problematic, and/or short on or contradicted by evidence.
[ 08. February 2004, 21:17: Message edited by: Rex Kerr ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 08. February 2004 22:41
Truman wrote quote:
What Rex needs to understand, is that the authors have chosen the kinds of organisms most likely able to produce novel cellular functions via neo-Darwinian processes: very small, rapidly reproducing creatures, possessing lots of junk DNA, able to survive very high mutation rates. I am merely accepting their chosen scenario and pointing out why this does not work once we insist on a minimum of biological realism.
Several points in response:
First, one can run the same experiment with an Ancestral genotype of just 22 instructions, eliminating the 'junk' instructions Truman thinks are necessary, and get the same general results as those reported in Lenski, et al.. If that's doubted, download the program and try it. The configuration files are easily edited. Data are much preferable to armchair speculations.
One can do the same with mutation rates edited to 1/10 of their default levels, allowing the simulation to run for 500,000 or 1,000,000 updates rather than the 100,000 Lenski, et al., used. As was pointed out earlier in this thread, the high mutation rate mostly serves to speed up the evolutionary process. One can run the same conditions with much lower mutation rates, and over more updates the same general outcomes occur.
Truman claims that he's pointing out "why this doesn't work." But it does "work" under those altered conditions. That's the beauty of an explicit research program where the methods are public and replicable: One can try it out for oneself. One needn't depend on armchair reasoning.
One or two other stray comments on Truman's paper, extracted from several pages of my notes:
1. Truman claimed that the avida model hasn't been "calibrated against real organisms." The model actually has been "calibrated" against real organisms and it has been used in comparative research against experiments on organic lifeforms. (See The Biology of Digital Organisms for examples.)
2. In the same paragraph, Truman claims ""Error catastrophe does not result, i.e., accumulation of flaws leading to extinction." In fact, in any given avida run, thousands of lineages go extinct. Again, try it and look at the data the program writes to disk.
Loss of functionality is often observed in one or another lineage. However, as evolution proceeds, digital organisms are competing against more and more reproductively fit peers, and loss of function leads to extinction of a lineage if it doesn't adapt successfully in the current competitive context. The lineage descended from a mutant that has lost a function will not persist in the avida world for long. It can persist for a while, but not forever unless the loss of function is due to what in retrospect is a "stage setting" mutation that enables acquisition of a more reproductively valuable function before the lineage completely fades out.
Throughout the paper, Truman seems to assume that the 3,600 critters in the avida world stay essentially identical to one another through time, that only one lineage is present at a given time. In fact, there may be several hundred lineages present in the population at any given time. Lineages come and go. Some persist because they have evolved sufficiently quickly to stay abreast of the competition as it becomes more stringent. They're the ones we see at the end of a run. But thousands have gone extinct in the process of getting to the end.
3. Truman claimed that "Statistically insignificant sequence space distances are assumed between novel, more complex functions."
Truman's discussion in this connection of the sparseness of protein families in protein space and of the improbability of finding a folded protein (Sauer reference) on pp 7-9 is mostly irrelevant, for three reasons. First, the Lenski, et al., paper at issue did not set out to show how all phenomena evolved. It had relatively modest goals, and it achieved them.
Second, the critique is contaminated by Truman's identification of instructions as codons (or perhaps amino acids), which leads to several non sequiturs and irrelevancies. The authors of the program identify avida instructions as more akin to genes, rather than to codons or amino acids, as Truman suggests..
Third (and most interesting), even if instructions are identified with codons, and the operation performed by an instruction is taken as analogous to protein folding/function, Truman's critique is based on the sparseness of the distribution of appropriately folded proteins in protein space. But that's irrelevant. What is relevant is the distribution of proteins in codon space. The distribution of proteins in protein space tells us nothing useful about how proteins are distributed in codon space - how they are distributed as a function of changes in codons. But that's the only distribution of interest. When I'm on the road and getting hungry, I'm not at all interested in the distribution of restaurants in size space; I'm interested in their distribution in geographical space. It's the graph induced by the mapping of codon changes into protein variants that is relevant.
RBH
[ 08. February 2004, 22:56: Message edited by: RBH ]
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 08. February 2004 22:53
About Sauer:
The following article appears to be at odds with the claims
quote:
A library of synthetic genes encoding 80- to 100-residue proteins composed mainly of random combinations of glutamine (Q), leucine (L), and arginine (R) has been expressed in Escherichia coli. These genes also encode an epitope tag and six carboxyl-terminal histidines. Screening of this library by immunoblotting showed that 5% of these QLR proteins are expressed at readily detectable levels. Three well-expressed QLR proteins were purified and characterized. Each of these proteins has significant -helical content, is largely resistant to degradation by Pronase, and has a distinct oligomeric structure. In addition, one protein unfolds in a highly cooperative manner. These properties of the QLR proteins demonstrate that they possess folded structures with some native-like properties. The QLR proteins differ from most natural proteins, however, in being remarkably resistant to denaturant-induced and thermal-induced unfolding and in being relatively insoluble in the absence of denaturants.
Folded Proteins Occur Frequently in Libraries of Random Amino Acid Sequences AR Davidson and RT Sauer Proceedings of the National Academy of Sciences, Vol 91, 2146-2150
Truman quotes an indirect reference for Sauer. Has Truman read the original papers?
[ 08. February 2004, 23:32: Message edited by: Pim van Meurs ]
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 29. March 2004 08:56
Reply to RBH posted 30. January 2004 00:51
RBH: The first sentence contains two* errors. First, Avida was not designed to "simulate" anything. Second, Avida has nothing to with "cellular networks," novel or otherwise.
RJT: I suspect RBH is implying I am making a stronger claim for what a “simulation” is than I intend. Certainly Avida is not a simulation of cellular processes in detail (e.g., kinetic details for the creation and regulation of chemical reactions, pathway /feedback inhibitaion details for regulation, protein degration, cell cycle, etc.)
Is it true Avida was not designed to “simulate” anything? Is it true that the authors are not deliberately conveying the impression Avida shows neo-Darwinian processes are plausibly able to generate the complex cellular / biological functions we observe?
Anyone reading the paper under discussion can determine that its purpose is to explain the origin of real, biological, complex organismal features in Darwinian terms. The well-known evolutionary biology professor (and author of a major university biology textbook) who brought this article to my attention, claimed it demonstrated that Darwinian processes suffice to produce novel metabolic processes. In fact, the specific topic under discussion between us was the origin of the Krebs Cycle.
The authors themselves wrote, “To examine the evolutionary origin of a complex feature in much greater detail than has previously been possible, we have performed experiments with digital organisms – computer programs” (p. 139). If the Avida model cannot be applied meaningfully to putative evolutionary processes for real biological systems, then it is worthless for the published intention.
Is it true Avida has nothing to [do] with “cellular networks”?. P. 139 “The next section describes our experimental system, including the logic functions that digital organisms can use to obtain energy.” Cellular networks, involving new enzymes, are needed to metabolize biochemicals. This is how energy is processed biologically: via integrated, regulated cellular networks. It is the origin of such networks which is under dispute: Intelligently Designed or not?
Further down the page, “Also, digital organisms compete for energy and, depending on the environment, can obtain energy by performing logical functions.” On p. 140, “However, an organism that evolved one or more of nine logic functions would obtain further energy” and, “Each mutation alters the genome and may change an organism’s phenotype, including its replication efficiency, computational metabolism and robustness.”
Replication and metabolim are regulated by finely tuned collular networks. I am simply stating the obvious in biological terms. At a conference this month, all the biologists familiar with Avida agreed the real-world analogy is to cellular networks: this is how energy is gained and processed from the environment.
RBH: Their goal is to provide a platform in which to do comparative research, contrasting and comparing the evolution of digital critters and biological critters, in order to study general principles of evolution; to test hypotheses about evolution; and (because of the massive data-gathering capabilities) to allow detailed analyses of the time course of evolving systems.
RJT: I’m trying to make the point that Avida differs fundamentally and qualitatively from biological reality. The evolutionary stepping stones are statistically easy to reach. The necessary genetic instructions to permit duplication in the real world must code for proteins: these must act physically to carry out the replication. Avida does not require all the genetic instructions to guide the physical hardward which carries out replication. Real, biological genetic system do! Mutations in Avida cannot destroy these (physically) necessary components, since these services are external to the Avida experiments themselves. These are provided for free.
I’m not addressing trivial matters, such as mutational rates which are too large. Until a model based on real biological facts has been designed in a semblance of a realistic manner, no general principles of evolution and certainly no detailed analysis of the time course is possible.
RBH: So Avida isn't a "simulation" of anything; it is an evolutionary system in itself. It shares some properties with biological systems and doesn't share others.
RJT: It is a computer based model which purports to demonstrate neo-Darwinian theory processes could produce novel complex functions. Virtually anyone reading the paper under discussion will assume the conclusions are applicable to the biological world and theory of evolution.
RBH: Avida, including the source code, is freely available on the Web. Rather than argue it, why not show it? The necessary resources are available.
RJT: My fundamental point is that it makes no sense to do this. The nature of computer logic functions, as used by Avida, defines how great the statistical “stepping stones” are. Although charming in a a cutsie manner, calibrating the structure and interactions of logic functions to real-world protein-based machines is not possible.
What proportion of random DNA sequences of the length of a small gene would provide advantages to an organisms? To be selectable, DNA must be both expressed and produce an acceptable proportion of mRNA (run-away production of mRNA or polypeptide would kill the organism). Based on the miniscule proportion of polypeptides able to even fold in a reliable and stable manner (one requirement for a protein), we know that the proportion of useful to non-useful DNA sequences is staggeringly small. This corresponds to the experiment the authors already attempted, (p. 143) in which the simplest function rewarded was EQU: NONE of the populations evolved EQU!
The make matters even more realistic, such as that the genetic translation and duplication apparatus must ALSO be coded for genetically, I would start with organisms having not 15 necessary instructions initially and 35 tandem, but perhaps 150 necessary and 5 tandem, to represent a more realistic proportion. If the author’s wildly optimistic settings and expensive computer cluster didn’t evolve EQU, I would be foolish to attempt computer runs vastly less likely to work.
RBH: The Nature paper of Lenski, et al, demonstrated a set of conditions in which evolutionary mechanisms generate products that meet Behe's definition of irreducible complexity, so one can no longer argue that those mechanisms are in principle not capable of doing so.
RJT: This is not correct. The 3,600 digital organisms have a very high likelihood of evolving according to the Avida conditions, which are biologically irrelevant: the stepping stones are probabilistically close together, the rewards for more complex functions immediate & very generous, and the proportion of “wasted” and destructive mutations unrealistically low. The fact that natural selection would favor the more rapidly reproducing variants was simply overlooked. This is one of my key points, which RBH has not addresssed.
RBH: Genome enlargement was not differentially rewarded; in the particular conditions of the Lenski, et al research, genome length was rendered selectively neutral by awarding SIPs in proportion to genome length.
RJT: Perhaps the authors should clear this up themselves for us. P. 143: “Each digital organism obtained ‘energy’ in the form of SIPs at a relative rate (standardized by the total demand of all organisms in the population) equal to the product of its genome length and computational merit, where the latter is the product of rewards for logic functions performed.”
It seems, that even in the absence of extra logic functions, the larger genomes “hog” more of the total energy available to be distributed among all members. Relatively less energy is left over for the little guys. And p. 140, “each organism receives SIPs in proportion to its genome length.”
I would be grateful for clarification of this point. In any event, organisms having to produce more genetic material during reproduction (and also more worthless polypeptide during evolutionary trials and errors) with the same number of logic functions would be disfavored in the real world: they would require more energy and material, and would reproduce more slowly each generation.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 29. March 2004 10:24
Repoly to RHB posted 30. January 2004 11:29
RBH created a fully evolved straw man with my name on it, then proceded to torch him. Excerpts:
RBH: “That is, Truman argues that because the experimental context is designed and controlled by humans, one can't generalize to biological organisms "in the wild." “
“Because humans control stuff like the energy supply to the Avida world and that world's 'physics' and 'chemistry,' we cannot learn anything interesting about a different world in which those aspects are "naturally" present.”
“Truman's critique must reduce to the observation that one can't generalize wholesale but must be aware of the limitations of a given experiment.”
All this is in reply to what I actually wrote:
RJT: Avida's organisms replicate thanks to external software and hardware which provide key services; the energy is supplied from external sources (e.g., electricity); their immediate environment is maintained and repaired by humans. Notice that these requirements cannot be damaged by mutations in the Avida model. However, free-living organisms need to have these critical functions encoded on their genes and these CAN be damaged by mutations.
My point is totally different. The Avida digital organisms can initially do everything necessary to survive, including replicate themselves.
In the biological world, organism replication and genetic material replication must be performed in a a real, physical manner, using protein machines. The instructions to create these physical, indispensible machines, are themselves coded genetically!
Furthermore, it is not enough to have some genetic instructions to code for the protein components: ribosomes, tRNA, dozens of enzymes… are also needed to provide indispensible services, and these must also be coded for on the genomes. Avida provides these for free: it does NOT code for the biologically necessary hardware!
That is one hidden reason why Avida “works”. P. 143: “The handwritten ancestral genome was 50 instructions long, of which 15 were required for efficient self-replications”. 15 instructions?!
Now we see why the unrealistically high mutations rates do not kill off all the Avida creatures within a few generations at most (‘error catastrophe’): only 15 instructions are exposed initially to this extinctional risk! A point mutation rate of 0.0025 per instructions copied (p. 143) leaves the necessary instructions unscathed almost every generation.
The genetic instructions needed to keep the infrastructure going, as found in living organisms, ARE exposed to mutational risk.
So, instead of 15 instructions, lets increase this a little bit (not too much, to keep computer runs reasonably short; we’ll just extrapolate our results).
Lets try runs using 3,600 organisms with 1,500 instructions needed for efficient self-replications and 35 unnecessary extra instructions for nature to play with (here an instructions corresponds roughly to a unique codon or amino acid). We retain the 0.0025 rate per instructions copied, and indels each with a probability of 0.05 (p. 143). At the same time, we must remove a couple of the intermediate stepping stones, to approach a little closer the real world situation. Our runs would now show many extinctions, and lots of great evolutionary progress (e.g., a newly evolved logical function) would be wiped a few generations later.
Of course, in the real world enough genetic material for just a handful of genes is unrealistic, so we repeat the work with 15,000 necessary instructions (0.0025 X 15,000 = ca. 38 point mutations per generation in the critically important portion). In these runs we’ll leave only one or two stepping stones to connect up to the EQU, to approach yet closer the real world statistics.
At this point we’ll probably decide the effort is not worth continuing to truly realistic biological scenarios. Avida only works thanks to biologically irrelevant parameter settings.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 29. March 2004 11:01
Responses to Charlie d. posted 30. January 2004 20:58
Charlie d.: However, AVIDA is certainly not meant to simulate "cellular networks" (did Truman really mean "molecular networks"…
RJT: My intention was intra-cellular. Good point, the wording could be confusing! Thanks. (I hope this is not part of BRH's discontent!)
Charlie d.: Also, as far as the rate of mutation is concerned, in addition to the point made above that reducing mutation rate would only have a quantitative, not qualitative effect on the outcome,
RJT: No, not "only". Ok, now. Rapidly reproducing organisms offer the most opportunities to generate variety via mutations. These tend to possess small genomes, which must have been the case 3 billion years ago if the evolutionary, Creator-less scenatio is true.
One of my key points was that such genomes would streamline: the members shedding unnecessary genetic material would reproduce more rapidly, and their ancestors and theirs. A streamlined genome results in the population, contra the Avida results, where this key insight is overlooked.
Slowing down the mutational rate favors the streamlining process: fewer opportunities to generate something useful would be provided among the superfluous genetic material, so more generations are provided in which genome truncation is the only action in town.
Charlie d.: I thought important to keep in mind out that high mutation rates in AVIDA are actually counterbalanced by the low population size within any single run - far below effective population sizes of biological organisms. In other words, most biological populations, due to their size, do in fact sample a dramatic amount of new genetic diversity in each generation, even if their mutation rates (at the individual level) are quite low.
RJT: Correct. But we introduce now a counteracting factor: random effects on survival and reproduction will wipe out countless “good” mutation long before they manage to fix in very large populations.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 29. March 2004 11:56
RJT: "In fact, I propose they were originally created with more information content than needed, to provide fault tolerance towards mutations. "
PIM: An interesting proposal, could you elaborate in some more detail that would help us evaluate this concept?
RJT: Suppose you coded the instructions for a guided missile to land within 2 millimeters of your bulls-eye. Something goes wrong (the target sways in the breeze) and the bomb lands 5 millimeters from where you wanted it too. Good enough, job done.
Dr. Schneider argues that proteins and binding sites co-evolved starting in both cases from random sequences. I consider this a hopeless task for natural selection acting on whole organisms, with new opportunities being provided by random mutations. There are too many ways to die which have nothing to do with a slightly improved binding site sequence.
Suppose the Intelligent Designer decided to create a robust system able to function in the future in spite of mutations. The information content (in Shanon’s sense) would permit identifying where to bind, and indeed be deliberately overspecified. The original created binding sites are placed where they belong, purposefully. (A precise sequence pattern would be
used, longer than necessary to identify the binding location unambiguously).
Random mutations can then alter the original binding sequence rapidly, unless a deadly mutation occurs. In this viewpoint mutations don’t generally evolve new and more complex functions, but devolve and degrade existing ones.
PIM: Schneider argues, and I tend to agree here, that "Many of Truman's complaints have to do with inessential points"
RJT: At this time he has not addressed any of my key points. He can also easily perform sensitivity analysis with his own software, and I explained exactly how.
I posed many questions. How does one justify the assumption that a dramatic selection advantage always occurs in the first generation starting from random protein and binding site sequences, for which no biological functional could possibly exist?
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 01. April 2004 00:57
I'm pressed for time these days, so I'll address just two of Truman's points for the moment, one technical, one more substantive. I hope to return to the rest next week sometime.
Selective neutrality of genotype length
The technical point has to do with the allocation of SIPs - 'energy' - to avida critters as a function of their genotype length. Interpreting the Lenski, et al., paper's description, Truman wrote quote:
Perhaps the authors should clear this up themselves for us. P. 143: "Each digital organism obtained 'energy' in the form of SIPs at a relative rate (standardized by the total demand of all organisms in the population) equal to the product of its genome length and computational merit, where the latter is the product of rewards for logic functions performed."
It seems, that even in the absence of extra logic functions, the larger genomes "hog" more of the total energy available to be distributed among all members. Relatively less energy is left over for the little guys. And p. 140, "each organism receives SIPs in proportion to its genome length."
As the last quotation plainly says, absent the performance of logic functions, critters received SIPs in proportion to their genotype length. On a per instruction basis, short genotypes and long genotypes received equivalent 'energy.' There is therefore no selective effect of genotype length as such. Short genotypes had the same amount of 'energy' on a per instruction basis as did long genotypes. The avida documentation has a more complete description of the various methods of allocating resources to critters that are available in the platform - see "Time Slicing" at that URL. Also see the discussion of Time Slicing in the avida Technical Manual that is included in the download of the program.
If a critter with a short genotype performs a logic function, its base allocation of SIPs (determined by its genotype length) is multiplied by the 'reward' for performing that function. If a critter with a long genotype performs the same logic function, its base allocation is similarly multipled by the reward factor for that function. The net effect is that on a per instruction basis, genotype length continues to be selectively neutral when critters earn extra 'energy' by performing logic functions. The longer genotype gets a larger absolute amount of 'energy,' but once again, equating base energy (before the multiplcation for tasks performed) on the basis of number of instructions in a genotype renders genotype length selectively neutral.
What (if anything) does avida "simulate"?
The second point that Truman makes is more substantive and is worth spending time on, so I'll address it fairly briefly here with the hope that I'll have time to come back to it later. It has to do with whether the avida platform "simulates" a specific biological structure or process. (IIRC, charlie d (I think) and I had an abbreviated version of this same discussion in the Literature Forum thread on this topic.)
Let me try to be clear about what I do and do not mean by saying that "Avida was not designed to 'simulate' anything."
The authors of the avida platform clearly intend it to be an environment in which hypotheses, conjectures, and claims about evolution in general can be tested. While in several places, in various papers describing the platform and research performed with it, they suggest rough parallels between features of avida and features of organic systems, they resist mapping specific features of avida and its critters onto specific features of organic systems. I will not give citations now - again, I'm pressed for time - but it seems clear to me in reading those papers that the creators of the avida platform did not intend to create a simulation of a specifically biological system, but rather intended to build a system in which evolution occurs independent of purely biological considerations.
That is, they took evolution at its most general level of analysis -- a population of imperferfect replicators with heritable variance, reproducing in an environment with limited resources (and therefore with competition and consequent selection on replicationally advantageous traits) -- and built a computer system within which evolution actually occurs at that level of analysis. It's not a "simulation" of evolution at that level, it instantiates evolution at that level. Somewhere (again, no cite, but I know it exists) they said that their intent in writing the avida system was to create a plaftorm for doing comparative research on evolutionary processes, where the comparison is between evolution occurring in the avida virtual world and evolution in the world of organic biology. And indeed, in The Biology of Digital Organisms, Wilke and Adami, two of the creators of avida, describe several examples of such comparisons.
The approach is then the study of evolutionary processes at one level of analysis above the specific instantiation - organic or digital. Since that line of research is in its infancy, one must be cautious. Nevertheless, there is much to be learned from doing comparative research.
Similarly, the Lenski, et al., paper takes a question in organic evolution, the evolution of complex systems, and studies it in an alternative evolutionary system, the avida world, which has the advantage of allowing one to record complete evolutionary histories and therefore do analyses that are not easy (or even possible) in an organic biology lab with E. coli in Petri dishes.
The results of the research are relevant to an issue raised in the Intelligent Design literature, namely whether "irreducibly complex" structures or processes can arise via the operation of evolutionary processes. "Irreducible complexity" is taken at the level of analysis in Behe's original definition, which is at a comparable level of analysis to that of the definition of evolution above: quote:
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.
Note that Behe's definition makes no more reference to evolution in organic systems than does the abstract conception of evolution I noted above. Indeed, Behe's iconic example of an irreducibly complex system is a mousetrap, so he clearly does not mean the concept to be confined to organic biological systems.
Irreducible complexity is at the core of of the Intelligent Design movement's claims about the unevolvability of certain kinds of systems, and is interpreted by ID proponents to be a signature of design by an Intelligent Agency. But if systems meeting the definition of irreducible complexity can evolve in a context that instantiates the general properties of evolutionary systems, then that claim is weakened, and indeed, is falsified with respect to the claim about evolutionary processes in general. And that's what the Lenski, et al., paper does: it demonstrates that systems meeting the definition of IC can evolve in a context that instantiates the general properties and processes of evolutionary systems.
The ID defense against that conclusion must then be something to the effect that evolution in organic systems is somehow incapable of doing what evolution in the avida world can do, that one cannot generalize from what is learned about evolutionary processes in the avida world to organic biology. That's a principal thrust of many of the defenses on this board and elsewhere, once one gets past the misapprehensions and misunderstandings of the avida platform itself that infect those defenses. In my next posting on this topic I may pose some problems or questions that assess understanding of the avida platform in order to sort out who understands it well enough to critique it and whose critiques are therefore worth addressing.
And that is all I have time for at the moment!
RBH
P.S. Added in late edit: I see that the brief discussion with charlie d about 'simulation' was in this thread, not that on the Literature Forum.
[ 01. April 2004, 14:56: Message edited by: RBH ]
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 08. April 2004 17:37
Reply to RBH’s post 30 Jan. 2004
[RBH]: “I started an Avida run using the configuration files of an old run in which mutations rates were set very low and Avida's analogue of lateral gene transfer was enabled.”
[Royal]: Why compensate the lower mutation rates with gratuitious free genes? How likely are these new “genes” to further the evolution of new functions according to the Avida program? Now compare this probability to real biological organisms: reading frames get scrambled, expression of existing genes often get messed up, etc.
[RBH]: “at 275,000 updates it has lineages that among them perform input-output operations that correspond to all 9 of the logic functions”
[Royal]: This is a step in the right direction. My analogy is an investment decision: consider all costs and all benefits, then determine whether a net loss or profit is to be expected. All the factors used by Avida are quantitatively absurd. Factories don’t cost 8 cents to build. So figuratively speaking you increased it to twelve. As more realism gets introduced into the parameter setting, and neglected realities are introduced, we’ll find that the net effect will cease to be development of novel, complex functions but rather degradation.
[RBH]: But it's already obvious that even with mutations rates one-tenth those used in the Lenski, et al, study, and with the analogue of lateral gene transfer enabled, those pesky irreducibly complex critters still evolve.
[Royal]: Correct. This is an inevitable consequence of the unrealistic model structure. For example, introduce now a penalty for increase in genome size (ceteris paribus) since in reality these lineages they would reproduce slower and require more energy and material (In the real world, extra genetic material has to be synthesized, which Avida neglects). The smaller organisms with the same value in terms of ‘logic functions’ would have better chances to obtain scarce resources. Without additional rewarded ‘logic functions’ these larger genomes would be at a selective disadvantage.
Secondly, increase the minimal amount of genetic material which is indispensible for the organism. The model must permit this to be destroyed also by mutations also. An autonomous, genetic replicator needs the instructions to produce the material which physically performs the replication work. In Avida the computer hardware and software does this risk-free.
IP: Logged
|
|
|
Printer-friendly view of this topic