Re: Haldane's Dilemma - clarifications - and Felsenstein [LONG]


"Malcolm" <regniztar@xxxxxxxxxxxxxx> wrote in message news:e7jp1h$2g2m$1@xxxxxxxxxxxxxxxxxxxxxx
Here we go.

You are putting a lot of work into this. I hope you won't take my
negative comments too, ummh..., negatively.

[snip]

One interesting thing to note is that, in this model, when the number of
genes is large - I origininally set it to 30000 to represent the number
thought to exist in the human genome, then the number of mutations that make
it into the population is approximately 2 * SELECTIONCOEFFICIENT. When I
tried to speed the program up by a factor of 300 by reducing the numebr of
genes to 100, this relationship disappeared.

Maybe this is worth pursuing.

This sounds like a consequence of the theory of 'nearly neutral' selection.
With lots of loci, the count of the number of improvements at any locus
rarely rises above +1 or falls below -1. With a low selection coefficient,
and a low population size, most loci are nearly neutral. But when the
number of loci falls to a small fraction of the number of generations, every
locus eventually accumulates a significant count of improvements, and
most loci are no longer nearly neutral.

The number 250000 represents the numebr of generations since the human /
chimp split, or as ReMine would say, the alleged human/chimp split.

If your goal here is to convince ReMine he is wrong about humans and chimps,
I suspect you are wasting your time and your computer's time. For one thing,
you would have to change your model (drastically) so that it more closely
matches hominid life histories. But if the goal is to understand whether the
rate of substitution is really limited in a model with fertility selection,
then I think that it can be done using a much smaller number of generations.

However this is only a model. it can be made better, for instance at the
moment males and females are absolutely symmetrial, whilst we know that in
human populations males have more variable fertility. So let's hope Walter
ReMine can, as Joe Felsenstein has done, suggest improvements to show the
effect he wants to show.

Don't hold your breath. Walter declined to respond to your model before,
and will probably do so again.

[snip]
/*
choose a parent
Params: candidates - list of potential parents
N - number in list
Returns: the selected parent
We take two at random, and return the one with the highest fitness,
or the second to be examined if equal.
*/
int choose(CREATURE *candidates, int N)
{
double target;
int top;
int bottom;
int mid;

target = uniform() * candidates[N-1].cumfitness;

bottom = 0;
top = N-1;
while(bottom != top && bottom != top-1)
{
mid = (top + bottom)/2;
if(candidates[mid].cumfitness < target)
bottom = mid;
else
top = mid;
}
if(candidates[mid].cumfitness < target)
mid++;

if(mid > 0 && candidates[mid-1].cumfitness > target)
mid--;

return mid;
}

I had a devil of a time figuring this out. It didn't help that the
block comment on this function is completely out of date with what
the function does now. When I finally did figure out what you were
doing, I had to admit that I found it rather clever. But it is not
the way I would have done it. Instead of keeping a cumulative
(running total) fitness for each individual in the population, I would
instead keep a single MAX fitness for the population. I would
randomly select a candidate parent as you used to do. The candidate
is tested with a probability of 'passing' being the ratio of his/her
fitness to the MAX fitness. If the candidate 'fails', try another
candidate until one passes. My approach is more efficient, but not
as clever as yours.

[snip]
for(i=0;i<NGENES;i++)
{
fitness += child->matchromosome[i];
fitness += child->patchromosome[i];
}

child->fitness = pow(1 + SELECTIONCOEFFICIENT, fitness);

This code snippet, like the out-of-date comments, illustrates how code
readability deteriorates over time. You are using the symbol 'fitness'
here with two very different meanings. One is an int count of
favorable mutations. The other is a double with a meaning more in
the Fisher-Wright tradition. Confusing to the reader.

[snip remainder]

Have you tried multiple runs, varying the mutation rate as I suggested?


.

Relevant Pages

  • Re: c interview
    ... undefined behaviour because printf is a varidac function. ... stdio.h before using printf and other headers as appropriate before ... Would make it a point that the size of int is assumption to be .... ...
    (comp.lang.c)
  • Re: Overlapping fitnesses
    ... > nature's baseline of morphological similarity. ... [snip some more] ... I've defined fitness! ... You haven't removed a thorn from the side of the ...
    (sci.bio.evolution)
  • Re: || putchar(ch == 177 ? ? : ch | 0100) == EOF)
    ... snip 160 lines of obsolete commentary ... void filecopy(FILE *ifp, FILE *ofp) ... int ch; ... Since you expect the file to contain control character, ...
    (comp.lang.c)
  • Re: stream io in c
    ... If you are using tabs to indent your code, consider switching to spaces, ... And note that 'c' is declared as an int, ... unsigned char. ...
    (comp.lang.c)
  • Re: "Relative fitness" and other evolution pertaining
    ... > fitness ... meagre mathematical models of biological evolution. ...
    (sci.bio.evolution)