A chance for input ... Dr. John Sanford

[sfs1]

If anyone in the scientific community saw this result, if they were prone to take his simulation seriously, it would all be over. No one is going to believe it comes close to simulating reality with these results - not without some major justification/explanation of these results and why they don't reflect on the correctness of the simulation at low values.

I'm in the scientific community, and I have even written (and published) genetic simulation software, as well as peer-reviewed others' programs. I would not treat this one kindly. A great deal of effort seems to have been spent on various flourishes, without any attempt to make the core of the model realistic.

As someone interested in creationism, I do find the program fascinating(*), though. What they are simulating is not an evolutionary biology model, but a model of how creationists think evolution works.

(*)Well, ok, mildly interesting. But I'm trying to avoid writing a referee report on a boring paper that I've already reviewed twice, and I've got the flu, so I have to take what I can get.

[Ansgar Seraph]

Still looking for those advanced parameters; might not find them in the 15 minutes I've got left at the computer.

I did find, however, the first paper Sanford presented at ICC 2008; I had some trouble tracking it down, due to what appears to be a mistype on the ICR web page:

Mendel’s Accountant: A New Population Genetics
Simulation Tool for Studying Mutation and Natural Selection

Hopefully, this will have some interesting information.

—Sam

[oxmixmudd]

That was my impression, too. The selection parameters don't seem to be doing their job. Here's the "Linkage" plot from my 90% run:

Attachment 65148

Even though there are more than 300,000 beneficial mutations than deleterious ones, the effect of the "near-neutral" deleterious mutations is dwarfing the beneficial block. I found an advanced setting where I could increase the "maximal beneficial effect per mutation" from the default (0.001, or 0.1% fitness increase) but I've got no idea how accurate that number is to start with.

Mendel's user manual says:

Mendel's Accountant User Manual, pg. 7

Maximal beneficial mutation effects –

A realistic upper limit must be placed upon beneficial mutations. This is because a single nucleotide change can expand total biological functionality of an organism only to a limited degree. The larger the genome and the greater the total genomic information, the less a single nucleotide is likely to increase the total. Researchers must make a judgment for themselves of what is a reasonable maximal value for a single base change. The MENDEL default value for this limit is 0.001. This limit implies that a single point mutation can increase total biological functionality by as much as 0.1%. In a genome such as man’s, assuming only 10% of the genome is functional, such a maximal impact point mutation might be viewed as equivalent to adding 300,000 new information-bearing base pairs each of which had the genome-wide average fitness contribution. Researchers need to honestly define the upper limit they feel is realistic for their species. However it should be obvious that, in all cases, the upper limit for beneficial mutation effects ought to correspond to a very small fraction of the total genomic information (i.e. a small number relative to one).

© source where applicable

So they could be hamstringing it both ways; using "hard" selection to make each deleterious mutation affect the population's fitness negatively and restraining beneficial mutations so that they have too small of an effect to matter in any realistic scenario.

—Sam

PS - Running the 95% Beneficial run now . . . it looks like fitness is hovering between .98 and .97 — it looks like a population needs 95 out of 100 mutations to be beneficial (at 0.001 maximal beneficial effect per mutation) to reach equilibrium (within 40,000 generations, at least).

I would say that is a big part of the problem. I would doubt that the true maximum potential benefit of a point mutation is that low - those the probability of a point mutation being more beneficial that that may be quite low. That is, I would say he at least needs to be able to assign a distribution to this parameter, where at some low probability, we may find a much larger benefit.

Also, there is no way to evaluate the combinatorial effect of groups of muttions. That is 1+1 may be quite a bit greater than 2. Indeed. mutations have been shown experimentally to behave like this:

Ma -> -.002
Mb -> -.001
Mc -> -.004

Ma+Mb+Mc -> +.01

Because in combination they result in some newer and better result, whereas alone they may damage what already is.

Jim

[rogue06]

Thankyou AS and sfs1 for the work you've put into this

[Ansgar Seraph]

Thankyou AS and sfs1 for the work you've put into this

I'm just the guy who managed to get the program running (I even sleep between runs) . . . all glory to sfs1!

Speaking of runs, I've been looking for those individual selection parameters you were talking about, sfs1, but I can't find 'em. I figured to show what the basic and advanced parameters were . . . if anyone spots what I'm supposed to be doing here, let me know and I'll run the test. If I get the x64 build of Windows 7 in the next few days, I'll borrow some RAM and try to run a population of 10,000 through 100,000 generations, too.

—Sam

[sfs1]

I'm just the guy who managed to get the program running (I even sleep between runs) . . . all glory to sfs1!

And I'm too lazy to get the program running. (Actually, if it were just a matter of compiling code and running it I'd try it on one of the 8 gig farm machines at work. But I don't want to be messing about with web server software anywhere, and especially not on work computers.)

Speaking of runs, I've been looking for those individual selection parameters you were talking about, sfs1, but I can't find 'em. I figured to show what the basic and advanced parameters were . . . if anyone spots what I'm supposed to be doing here, let me know and I'll run the test.

Parameter 5 under "mutations" is one of them: "Special case - Consider all mutations equal". That must be the one that gives all mutations the same selection coefficient. I can't see where you're supposed to set that value, however, so that may not help much.

Another interesting parameter is 4 under "selection", "Fertility declining with fitness". Unchecking that presumably switches from hard selection to soft.

Finally, I note that the default engine seems to be written in Fortran. I'd bet quite a bit that means it was written by Baumgardner or another physics computational guy, and probably an older one at that -- speaking as an increasingly old ex-physics computational person who used to use Fortran myself. You very seldom see biologists writing in Fortran. (Looking at the author list, yeah, I see Baumgardner is listed as the main developer.)

[Ansgar Seraph]

Parameter 5 under "mutations" is one of them: "Special case - Consider all mutations equal". That must be the one that gives all mutations the same selection coefficient. I can't see where you're supposed to set that value, however, so that may not help much.

OK, I'll set that up. Here's the description from the User Manual:

Mendel's Accountant User Manual

Special case: All mutations are given an identical fitness effect.

It is sometimes of theoretical interest to model what would happen if all deleterious mutations had an equal effect and all beneficials had an equal effect. If this option is selected, one needs to specify those constant values for both deleterious and beneficial mutations. When this special case is chosen, it overrides the actual mutation distribution parameters otherwise specified. Selection is consequently based essentially upon each individual’s total mutation count.

© source where applicable

Another interesting parameter is 4 under "selection", "Fertility declining with fitness". Unchecking that presumably switches from hard selection to soft.

I think this parameter is to keep the population at, say 10,000, regardless of overall fitness; the fitness still declines for every deleterious mutation but, unlike my mouse population, the number of organisms does not decline and speed up the population collapse:

Mendel's Accountant User Manual

Fertility declines as fitness declines.

It is widely recognized that when fitness declines, fertility also declines. This in turn affects population surplus, which affects selection efficiency, and can eventually result in “mutational meltdown”. To model this, we have included an option wherein fertility declines proportional to the square of the fitness decline. The resulting fertility decline is initially very subtle, but becomes increasingly severe as fitness approaches zero. The default value is “Yes”, which means that fertility declines with fitness, especially as fitness approaches zero.

© source where applicable

They explain it a bit more in "Using Numerical Simulation to Test the Validity of Neo-Darwinian Theory":

Using Numerical Simulation to Test the Validity of Neo-Darwinian Theory", Sanford et al., Sixth International Conference on Creationism, 2008

There is another especially important variable which affects the time to extinction, and counteracts the effect of the near-extinction partial truncation selection. As a population degenerates, there is one biological variable which can not be kept constant— fertility. It is well known that declining genetic fitness causes declining fertility. But any reduction in fertility will accelerate the rate of fitness decline—creating an accelerating downward cycle. Declining fertility reduces population surplus, which in turn reduces selection intensity (simply because there are fewer surplus individuals that can be “selected away”). This in turn causes still more rapid mutation accumulation. This causes faster fitness decline, and this reduces fertility still further. The vicious cycle begins to accelerate—causing what is called mutational meltdown.

© source where applicable

Thanks for the guidance; I'll set up a run in a minute here and hopefully post up before I've got to head out for the afternoon.

—Sam

[Ansgar Seraph]

Mendel's Accountant just gives me a blank page when the option to treat all mutations as equal is selected. I'm taking a friend back home today, so I'll have free time tonight to hopefully set up a Linux build, drop more RAM into the system and have a better system on which to run Mendel.

Don't wait up, though, y'all . . . configuring Linux has never been a strong suit of mine

[sfs1]

Mendel's Accountant User Manual

Special case: All mutations are given an identical fitness effect.

If this option is selected, one needs to specify those constant values for both deleterious and beneficial mutations.

© source where applicable

That's the part I didn't see how to do: how to specify the constant values. Is it obvious?

I think this parameter is to keep the population at, say 10,000, regardless of overall fitness; the fitness still declines for every deleterious mutation but, unlike my mouse population, the number of organisms does not decline and speed up the population collapse:

That's precisely the distinction between soft and hard selection. With this switch off, fitness only affects the relative probability of each individual reproducing; it doesn't do anything to the health of the population as a whole. Assuming they calculate the loss of fitness as a proportion of existing fitness, then the population shouldn't go extinct for a very long time -- basically not until the fitness gets too small to be represented by whatever floating point type they're using.

[Ansgar Seraph]

That's the part I didn't see how to do: how to specify the constant values. Is it obvious?

Seemingly! When you check the box, two form boxes pop up under it. The input value for deleterious mutations is a bit of bad code; I can see the code to define what the input does in place of the actual input box and when I change the beneficial input box, the program hangs.

That's precisely the distinction between soft and hard selection. With this switch off, fitness only affects the relative probability of each individual reproducing; it doesn't do anything to the health of the population as a whole. Assuming they calculate the loss of fitness as a proportion of existing fitness, then the population shouldn't go extinct for a very long time -- basically not until the fitness gets too small to be represented by whatever floating point type they're using.

I'll do a couple runs to verify once I've got the ol' Linux up n' running. Thanks for the guiding hand!

—Sam

[Ansgar Seraph]

Got Linux set up, added 1GB of RAM . . . and discovered that there is no listed Linux distribution package for Mendel's Accountant on SourceForge, despite it being prominently mentioned on MENDEL's home page. Displeasing.

So I might have to wait on the big population stuff until my x64 Windows 7 arrives. I'll run the special equal-mutation project again. I got it working in MENDEL 1.2.1 but it crashed the population in less than 300 generations (at 0.001 for both deleterious and beneficial mutations and beneficial mutations at 10%).

--Sam

[Ansgar Seraph]

I ran the "Special Case: Consider All Mutations Equal" run twice, once with "Fertility Declining with Fitness" checked and once with it unchecked. In both runs, mutations were considered to have an equal effect of 0.001.

Beneficial Mutation Rate: 0.1 (10%)

Population: 1000

Max. Generations: 6000

This is the run with fertility declining with fitness:

Attachment 65206 Attachment 65205

The run only lasted for 292 generations, with a final fitness of 0.11, deleterious mutations totaling 2144 and beneficial mutations totaling 364.

Sorry I can't post more graphs; I have to use MENDEL 1.2.1 for this, since the parameter is broken in 1.4.1.

Second run posted up in justasec.

—Sam

[Ansgar Seraph]

This is the run with fertility independent of fitness:

Attachment 65209 Attachment 65210

You can see that this run drops the fitness nearly to where the last population dies . . . but it seems to have turned the corner just in time and is slowly advancing in fitness. I wondered why the population died after 789 generations. At the end of the output file, MENDEL presented the error, "Favorable mutation count exceeds limit". It seems that MENDEL is unable to process favorable mutations over 1788 (the number of beneficial mutations in the final generation).

In the end, the population had a fitness of 0.2297 (and rising), with 3329 deleterious mutations and 1788 beneficial mutations.

—Sam


—Sam[/quote]

[Dr.GH]

So, it does seem that this creationist "model" has the result predetermined.

[Ansgar Seraph]

So, it does seem that this creationist "model" has the result predetermined.

I think so, too.

Here's the deal I made last night, the "Soft Selection" run:


Beneficial Mutation Rate: 0.001
Population: 1000
Max Generations: 25,000
Advanced Setting: Fertility Declining with Fitness: Unchecked

Attachment 65221 Attachment 65222


Even with soft selection turned on and a higher-than-default number of beneficial mutations, the population only lasted for 22989 generations. If these were mice, that would be 2874 years (humans would be ~750,000, assuming a 25 year generational gap). Just another reason that Sanford ought to explain why a population of 1,000 is biologically realistic.

I would like to run the program with a population of 10,000 for something like 50,000 generations but the program won't start something that big (it looks like that could take anywhere from 8GB-16GB RAM).

—Sam