445 thoughts on “Evolution Visualized”
Leave a Reply
You must be logged in to post a comment.
The Skeptical Zone
"I beseech you, in the bowels of Christ, think it possible that you may be mistaken."
You must be logged in to post a comment.
JoeCoder,
Your description conflicts with the code. According to the code, there are three selection schemes:
1) Unrestricted probability selection;
2) Strict proportionality probability selection; and
3) Partial truncation selection.
All three use truncation selection — it isn’t a separate mode. The differences among the three concern how the phenotypic fitnesses are modified, not whether truncation selection is used. Also, both #1 and #2 divide by random numbers. #3 divides by a more complex expression that includes a random number. Truncation selection and division by random numbers are not mutually exclusive, contrary to your statement.
No, all three give rates. You just have to read the entire paper and look at the tables.
The last one gives estimates drawn from other studies, for the duplication rate in humans, in table 1, measured at three loci. They give:
The rates vary between species, with bacteria apparently having extremely high rates. But overall, the rate of duplications is several orders of magnitude greater than the rate of point mutations pr nucleotide for all species for which we have estimates.
Yes, because there’s >3 billion nucleotides in the human genome, but “only” ~22.000 protein coding genes.
Here’s the crucial aspect:
But duplications are not restricted in size to just one gene/pr duplication. They span a range in size from minor segments of incomplete genes a few hundred bases in size, to huge chunks encompassing dusins of genes over half a megabase in size in total. It’s hard to give an average estimate because different methods give different estimates and they also seem to differ across loci, species and local conditions. Over 14 loci measured in C elegans, a typical size was ~3 kb. Oddly enough, when the rate of adaptation is high the size of fixed duplications seems to go way above the average duplication size (2nd reference, fig 3.)
Yes, why not? We’re getting better at that all the time.
Could you give some references? What is even nutty about that? They mutate, the mutations they suffer are random, so by chance they happen to affect some Alu’s such that they start being active again, and not others. What’s the problem with this?
What do you mean by punctuation of transposition mutations?
What is a punctuated mutation rate?
You mean we can’t easily model something that seems to randomly vary to a large extend over great timescales? Stop the press.
Cataclysmic? Is that an objective measure of their effect or just rhetoric?
Have you checked?
Took me literally 30 seconds of Pubmed to find this: Sequence conservation in Alu evolution.
“Despite variability, all Alu elements can be related to an average consensus sequence, as done by Kariya et al. (9) from an alignment of 50 Alu sequences (for a review see ref.3). Recently, Jurka and Smith (10; cf. also 11-14) were able to classify Alu elements into families, J, Sa, Sc and Sb, based on the presence of correlated nucleotide substitutions in a number of sequence positions. The age of these families is in the above order, as judged both by their mutational diversity and their divergence from the 7SL RNA sequence.“
Apparently there’s a pretty clear pattern of diversification of the Alu subfamilies from an ancestral 7SL RNA sequence over time. Why is the subsequent diversification within the J-subfamily following it’s initial birth, of particular importance to you?
Does what, specifically, agree with molecular clock predictions? What molecular clock predictions specifically?
Here’s a study on some of that stuff “no one looks at”, for the Ya-subfamily:
Analysis of the human Alu Ya-lineage.
Here’s one for the Yb-subfamily: Genome-wide analysis of the human Alu Yb-lineage.
Here’s one for the Yd-subfamily Comprehensive analysis of two Alu Yd subfamilies.
Here’s one for the Ye-subfamily:
Analysis of the human Alu Ye lineage.
But clearly “no one looks at this stuff”. You want one on the J-subfamily and you want it now and if you can’t get that, then “no one looks at this stuff” and it’s probably because they know it falsifies evolution but they just hate God too much to reveal that. Right?
Modeling what, exactly? I believe that Tom’s point is that the model that the software is supposed to be implementing is not clearly defined anywhere. If we had that it would be sufficient to re-implement MA and compare results.
Do you have any evidence from biology to support that claim?
Cite, please?
Usually the spam filter only blocks comments with a large number of links, which an admin can then review and approve. On rare occasion I’ve had a comment fail to post, but I’ve always gotten an error page when that happens. Has your comment reappeared?
Well, a concrete real-world experiment seems to contradict that an increased mutation rate necessarily comes with a fitness decline. The mutator phenotype has arisen on multiple occasions in the LTEE without negatively affecting the rate of adaptation compared to lineages that do not have a mutator phenotype. So apparently something is going on in the LTEE that is not captured in MA.
No, actually mammals have a lower mutation rate pr nucleotide, but they have much much larger genomes than bacteria, so they suffer many more mutations pr generation simply because of the size of the genome.
Then there’s the fact that the fraction of the genome that is functional is in dispute between us. How much of the genome does MA predict should be junk for the mutation rate to not lead to mutational meltdown for homo sapiens?
The mutation rate thing is a canard. Rates vary all over the place, and mammals are not special. (I might interject that creationists have some odd love of talking about “higher” species of animal, and get particularly nervous about human evolution and human lineage. When is the last time you heard a creationist specializing in plant evolution, or fungus evolution?)
There’s some mention of viruses and bacteria going int genetic meltdown when confronted with toxic pharmaceuticals. And yet none of these microbes have gone extinct due to meltdown.
Rather important if the purpose of your model is to disprove physics. (and the purpose of MA is to support YEC).
And yes, physics is correct about the age of the earth, and biology is correct about the continuous lineage of living things. So MA needs to be modified until it comports with reality.
It’s the model that needs correcting. I will stipulate to its doing the arithmetic correctly.
Wouldn’t it be simpler just to exclude that portion of the genome which is known to be insensitive to mutations? At least as a reasonable approximation.
And does MA treat protein coding and regulation as equally sensitive to mutation? Equally subject to purifying selection?
I found 4 comments of yours that were flagged as spam. I have “despammed” them. They should be there now. But they will show up in the original order, so people might not see them if they only look at new comments.
Pro tip for this website: put links in separate posts.
That’s true of uncommondescent also. It’s a WordPress thing.
Hello kieths!
There are four modes, not three. Truncation selection is only done when (selection_scheme == 1), which does not execute any of the randomization branches. Are you assuming they use truncation selection because they sort the population by fitness*randomnum() and then remove (truncate) those with the lowest fitness? How would you recommend removing those with lower fitness? They have to be removed somehow. To implement unrestricted probability selection I think fitness should be multipled by randomnum^2. Changing my javascript test to use the square also gives Barney Fife and Genghis Khan the correct ratio. Although it’s possible I’m missing something.
However, using truncation selection and this less-random-than-it-should-be version of unrestricted probability selection only make selection in Mendel MORE efficient. Changing these things would only accelerate the rate of deleterious accumulation.
Hello Rumraket. Yesterday you wrote:
Today you write:
I don’t think very many people would interpret your first statement to mean your second.
That’s equivalent to I said. I am beginning to think that you deliberately buy 12 loaves of bread just so to start an argument 😛
The Mendel authors report: “With an average of 1 new mutation distributed in a Poisson manner and with four of every six offspring selectively eliminated, truncation selection is able to exclude every offspring that has one or more mutations.”
But truncation selection is unrealistically stronger than what natural selection can be. I did my own run of Mendel for 5000 generations, 8 offspring per female (reproductive rate=4) and a mutation rate of 0.5:
The population’s goes down faster at first but then begins to slow. I would’ve done more generations to see if it levelsoff completely, but this run already took half an hour.
e coli get about one mutation every 200 to 2000 generations. The mutator phenotypes in the LTEE increased the mutation rate by 1 and 2 orders of magnitude, depending on the lineage. That’s still a lot less than one mutation per generation. Selection is also much much stronger in bacteria than in mammals. And they still lost half a dozen genes, including every lineage losing the RBS operon.
Hey Patrick. Thanks for the comments.
Human population genetics in general. If you can point me to a simulation that takes into account as many relevant parameters as Mendel, we can give it a try and see what it does.
This Nature paper says “humans are carrying around larger numbers of deleterious mutations than they did a few thousand years ago”. I’m not sure if that’s what you’re looking for. You could also suppose that any time the load increases it’s due to relaxed selection.
Casey Luskin talks about Springer’s rejection of the Biological Information volume after it had passed peer review. If you have another version of these events you’d like to share, I’ll give it a read.
When I replied to Tom I broke it into several parts to figure out which piece was causing trouble. It was a link back to a previous comment. I stopped putting it inside an <a> tag and then it let me post it.
Hey petrushka!
Without a rate at least as much as the human rate, how do you get enough neutral differences in mammal genomes during the time available?
Ironically John Sanford’s background is in plant genetics.
With their low mutation rates, why would we expect them to? We know that “mutations must be kept at a very low level to maintain genome integrity and yet must be frequent enough to support evolutionary change”
I already responded to you when you made this objection a couple pages ago. Maybe it was caught in the spam filter? Page 4, comment 34.
Mendel already does this implicitly by asking for the “non neutral mutation rate” rather than the total mutation rate. That is why we are using mutation rates like 0.5, 10, or 20. Instead of 100.
Thansk Neil for approving my comments. This has likely contributed to my frustration of having to answer the same questions more than once. So I apologize to everyone if I was snarky.
From your link:
Perhaps in 90,00 more generations, we’re all gonna die.
I remember seeing this part and it doesn’t make sense to me. Why would increased del load not lead to more disease? Maybe is he talking about buffering through redundancy or something else? Regadless, the data shows an increase in the number of del. mutations carried.
JoeCoder,
No, it’s done for all of them:
if (selection_scheme <= 4) {
/* Apply truncation selection to reduce the population size to
current_pop_size. */
Tournament selection and roulette wheel selection are common alternatives to truncation.
I mean, what is the specific model that MA implements?
The number of parameters doesn’t matter if the underlying model doesn’t reflect biological reality. Again, it appears that MA is smoke and mirrors designed to impress the rubes in the pews.
JoeCoder,
Your Javascript program only implements part of the MA probability selection algorithm.
What you’re testing is not how often each survives truncation, but rather how often Genghis’s fitness remains greater than Barney’s after each has been divided by a separate random number from the interval (0,1).
Again, this is about ordering, not survival.
As for the 10% vs. 20% issue, you’re being misled by your intuition. Here’s what’s really going on:
Let Fg represent Genghis’s pre-adjustment fitness and Fb Barney’s. Rg will be the random number by which Genghis’s fitness is multiplied* and Rb will fill that role for Barney.
If you do a little algebra, you’ll see that Barney’s adjusted fitness will surpass Barney’s as long as Rb/Rg is greater than Fg/Fb.
Rb/Rg is a ratio of two uniform random variables sampled from the interval (0,1).
The probability that the ratio Rb/Rg is greater than Fg/Fb is equal to (Fb/Fg)/2. Plugging in the fitnesses, we get a probability of .454545…. for a Barney “victory” and .545454… for Genghis. The ratio of those two equals 1.0/1.2, not the 1.0/1.1 that you were expecting.
* I’m using multiplication instead of division, as it is more intuitive and does not affect the result.
Thanks for your response, but I didn’t not describe the problem as well as I could, and had read references about the J family even better than the ones you provided.. So let me restate the issue.
There is divergence in the J family such that if the molecular clock hypothesis is correct, it indicates the J family is about 44 million years old by looking at the sequence divergence in the family members residing in the genome of humans. I depict the graph showing this and the 11% divergence within the same genome (intra species).
The problem however is that this divergence is measured within the Alus in the same genome. If we do interspecies divergence however, we may come up with a paradox.
If the same J family members appear in each primate line long after their supposed split, this would require convergent evolution of mutations in the corresponding Alus in the same genomic locations in the other primate lines.
The problem is analogous to the problem of SINES in rats and rhodents that Sternberg reported on:
Did the same point mutations simultaneously appear in J-Alus within gibbons and humans in the same genomic locations long after the supposed split. If the same Alu J family members are in the same genomic locations in other primates long after their split, this means that somehow magically the Alu J members mutated the same way in each of the primate lineages. That’s the problem that should be looked at. I don’t get the feeling anyone really wants to probe this question. And even if the J families in other primates are different but at the same genomic locations, it results in the same problem that Sternberg found for the SINEs for rats and mice.
The graph below is from here:
http://genome.cshlp.org/content/14/11/2245.long
Hey Patrick,
If that’s their goal they’re doing a very lousy job at it. If probability selection divided by randnum^2 as it should, then selection in Mendel would be even weaker and deleterious mutations would accumulate even faster.
Let’s get to the meat of the matter: You seem to think Mendel is suspect because it shows declining fitness when mutation rates are high? Take a look at the formula given earlier by Dr. F (page 3, comment 22):
A reproductive excess of 1.718 means 2.718 offspring per person, or 5.436 per mother. This formula simplifies to offspering_per_mother = 2e^u. For u=10, that means each person would need at least 44 thousand offspring. Now in my reply to Dr. F. I argue that the limit might be a little higher than 1 or 2. But at the same time this unrealistically assumes all death is due to selection. So weight these two together I expect it’s a good approximation.
Dan Graur also accepts the implications of this formula[1]. So does Larry Moran: “if the deleterious mutation rate is too high, the species will go extinct… It should be no more than 1 or 2 deleterious mutations per generation.”[2]. Ohno estimated that at most 6% could be functional (supposing a lot of duplicate genes to counter the effect) or else load would be too high.[3]
Given these calculations by Ohno, Moran, Graur, and Dr. F., a deleterious mutation rate of u=10 should definitely not be tolerable. But when the Mendel authors show declining fitness for u=10 (the value used in their papers), it’s untrustworthy and designed to trick gullible, dumb Christians?
Sources in the following comment, if they’re not eaten by the span filter:
Even Salvador agrees with me!
[1] http://sandwalk.blogspot.com/2015/01/a-lesson-on-genetic-load.html
[2] http://sandwalk.blogspot.com/2014/04/a-creationist-tries-to-understand.html
[3] http://www.junkdna.com/ohno.html (page 367)
Hey Keiths,
The specific algorithm performed following that comment is truncation selection. But the selection scheme as a whole is not truncation selection unless selection_scheme==4, because selection_schemes 1-3 apply randomization (lines 474-532) and strict truncation selection uses no randomization.
I agree that the comment could be more clear as to what’s going on.
I think you mean that Ghengis’s will surpass Barney’s?
To get a ratio of 1.1 you have to divide (or multiply) by randomnum()*randomnum(), instead of just by randomnum(). I think if Mendel did the same it would accurately model probability selection instead of making selection too strong. Is this also what you’re saying? I admit I’m not sure.
Joe referenced this:
Dang right I’m correct. Larry’s own figure:
Despite this Larry says:
Like how? I gave figures approximately the same as Larry of about 1 bad mutation per generation per individual. The 0.5 figure came from some of Muller’s writings. And even my figure is qualitatively in agreement with Larry’s.
Graur also gave his famous “bonkers” comment based on application of the Poisson distribution which I elaborated here with math derivations:
In Gruar’s own words:
Ironically, I actually agree with his calculations. As I said, ironically creationists like myself and John Sanford agree with the mutational load arguments of Muller, Joe Felsenstein and Dan Graur, but for different reasons.
Graur summarized the problem well, “If ENOCDE is right, evolution is wrong”. I agree. What’s the problem?
Petushka you wrote this above and I missed it earlier:
If high mutation rates are not a problem, why is it that no microbes with human-like mutation rates exist and continue to exist naturally? The only way to get microbes with mutation rates approaching that of humans is to use mutagens, and then they go extinct.
If you’re talking about genetic algorithms. But what about in population genetics models? That would be more relevant.
No, I think it’s suspect because
a) the model of population genetics that it implements is not clearly documented anywhere and
b) none of the results have been validated against reality.
Document the model in sufficient detail to allow a clean room implementation and then we’ll see if all the spinning wheels and flashing lights are more than a distraction.
Fixed that for you.
Patrick, as of yesterday I have now agreed that the implementation of Mendel should be better documented. However as Sal and I have shared the result of deleterious accumulation is concordant with what we should expect to happen. We also lack a better tool than Mendel to simulate human population genetics.
Do you agree or disagree that u=10 is an intolerable mutation rate? If you disagree, on what grounds, and what do you think is tolerable?
You’re still not addressing the issue. Diving into minutia of the implementation is a distraction. The important question is (h/t Tom English): What is the exact model that MA is implementing, in enough detail to allow an alternative implementation to be created? This is followed immediately by: What real world observations support the results of that model?
Agreed with the rhetorical objection, but it avoids another problem.
They don’t go extinct because the S-coefficients renormalize (for lack of a better word).
To illustrate.
Suppose we are modeling a deleterious trait like winglessness or blindness in a population. The S-coefficient is negative. Well suppose some strange set of circumstances induces a fish to go blind because it ends up in a cave or a beetle go wingless (for whatever reason) and further such deleterious traits get fixed in the population. What happens to the S-coefficient of these formerly deleterious traits now that the traits are fixed.
Well, S goes to 1 (blindness becomes the new normal, winglessness becomes the new normal), and hence the deleterious trait is now no longer deleterious thanks to definitional maneuvering, not because the defect (blindess or winglessness) was ever repaired. It got “fixed” in the pop gen sense, but not in the medical or engineering sense. Solution to the mutational problem via equivocation, not actual functional correction.
So that’s why they don’t go extinct, they just go on living with defects.
Eh, mendelian inheritance.
But this would all be moot if the Mendel detractors provided the software suite of their choice to model these issues. We can then run the sims with that software.
There is the software of Jody Hey’s lab. Is there anything more user friendly out there than his stuff? One of the Mendel Architects, Water ReMine, thought Hey’s stuff was good.
Hey Patrick,
The main issue is whether we go extinct from deleterious load. That’s where this all started from with my comment #1 on page 3. MA is just one tool used to show that this will happen. Can you answer my question about u=10?
MA implements its own algorithm. It can be reproduced by reading the source but I agree this is less than ideal.
Humans have more deleterious mutations than we used to. A high mutation rate in microbes makes them go extinct.
Well, there’s the earlier issue of whether or not any such thing exists in reality.
Without a clear description of the model that MA implements, it is not possible to support that statement.
Can you tell me how many faeries are dancing at the foot of the garden? You’re focusing on the bells and whistles of a piece of software without any understanding of what it’s actually doing. Without a clear explication of the model, all the fancy features are meaningless.
Rumraket asked:
From this paper:
Although the Alus is only about 300bp long, there are about 1 million copies of them and each subfamily has tens of thousands of copies.
So the problem is a mutation supposedly appears in an Alu, and then it propogates like crazy to make thousands of copies in the genome. If it happened in on one species line, maybe it’s not a problem for evolutionary theory, but if it happens in multiple species lines in the same genomic locations after a supposed split of primate lineages, it becomes a serious problem.
The problem is how these punctuated propagations end up in the same genomic locations of primate lineages after they supposedly split. It’s the Sternberg paradox.
The U concept does not come out of Mendel, it comes out of pop gen literature, for that matter a basic probability textbook!
https://en.wikipedia.org/wiki/Poisson_distribution
The symbols are different in the wiki article than what are used in pop gen literature, so I’ll clarify which symbols are what.
Substitute U for lambda and let k = 0, and you get what Joe is referring to.
It’s the same concept that is seen many places like the here:
I documented the explanation right here at the bottom:
It’s probably not mentioned in Mendel documentation since this stuff is so elementary and is expected knowledge of practitioners of pop gen. It’s not like we should expect Mendel documentation to spoon feed readers about the Poisson distribution’s relevance to population genetics.
I’m not talking about the bits and pieces of the software that creationists apparently find so distracting. I’m asking for a full, detailed description of the model that MA implements. That would allow for review by our resident biologists and permit creation of an alternative implementation should that be of interest.
Right now all we’re seeing is the equivalent of a black box with blinkenlights.
JoeCoder:
No, selection scheme 4 is what they call partial truncation selection. All four schemes use truncation selection.
No, partial truncation selection does use randomization. It’s just a different randomization scheme from the others.
It’s right there in the code, JoeCoder.
keiths, I should have been less ambiguous. When I say “strict truncation selection”or even just “truncation selection” I do not mean partial truncation selection. Partial truncation selection uses some randomness.
Unrestricted probability selection does not use any form of truncation selection. The same truncation algorithm is applied to it as with truncation selection (line 533), but having applied randomization to it on line 485 prevents it from being actual truncation selection. I agree that the comment on line 533 should explain this better.
Regardless, even when truncation selection is used, it makes selection much more efficient. So this gives us no help in resolving the mutational load problem, or showing that reality would be better at dealing with it than Mendel.
JoeCoder:
You’re confused, JoeCoder. You wrote:
That’s incorrect because selection scheme 4 is partial truncation selection, and it does use randomization.
I’m afraid this discussion won’t be very productive if you can’t read and interpret code, JoeCoder.
The U in question wasn’t specific to Mendel, but a known concept in popgen.
Some possible answers to JoeCoder’s question:
1. yes
2. no
3. I don’t know enough to have an opinion, so I can’t at this time
The response offered however was:
The U concept is not specific to Mendel.
FWIW, I don’t reference the results of Mendel much in these discussions because it was written by creationists, and even if were right, just because of guilt by association, it will be rejected. By way of contrast, the original WEASEL is long lost, no documentation, but it is celebrated as Darwin’s truth.
So can we provide documentation for Mendel’s Accountant? Probably not.
You don’t need Mendel to be documented to build your own pop gen software. One could build something for example that should yield Kimura’s famous equations. Mendel already models it correctly, and TSZ developers should be able to build one that shows neutral mutation rate equals neutral fixation rate. If the software doesn’t generate expected results then that suggests the software has issues.
Next, one could write something to model the Muller limit as I described here:
If the population (with reproductive capacity like humans) doesn’t go into meltdown when U=10, something is wrong with the software. Muller gave a figure of 0.5 or 1.0, you can try U=3. According to pop gen literature that should definitely precipitate meltdown unless there is some mechanism like synergistic epistasis (or whatever).
This is doable and a lot more relevant than WEASEL.
The other things that can be confirmed are Ohta and Kimura’s prediction about fixation of near neutrals. Mendel models that and is in agreement.
That’s the bulk of Mendel’s important results which agree with known and accepted and published literature.
But we don’t need to go there and verify Mendel when these questions can be independently explored right here without reference to Mendel.
So what will happen of the TSZ pop gen software computes the Muller limit to be 1 or 2 bad mutations per individual per generation when it appears ENCODE research would suggest humans have over 40 bad mutations per individual per generation. Will everyone then say, “If ENCODE is right, evolution is wrong” ?
kieths, my degree is in computer science and I’ve written code every day on a full time basis for 15 years now, using about a dozen different programming languages. I understand the code, but I didn’t have it in front of me when I wrote that comment and I mis-remembered what number mapped to what mode. Selection scheme 1 is the one that uses strict truncation selection.
Well if all the individual in the population has a defect, strict truncation selection would kill everyone. End of evolution.
It looks to me like mutations per base pair is similar in all cellular organisms. Somewhat higher in DNA viruses and a lot higher in RNA viruses.