Text 1285, 144 rader
Skriven 2005-01-01 06:01:00 av Perplexed In Peoria (1:278/230)
Ärende: Re: The "fuel" of evoluti
=================================



"Tim Tyler" <tim@tt1lock.org> wrote in message
news:cr460a$2hct$1@darwin.ediacara.org...
> Perplexed in Peoria <jimmenegay@sbcglobal.net> wrote or quoted:
> > "Tim Tyler" <tim@tt1lock.org> wrote in message
news:cquoj6$u8q$1@darwin.ediacara.org...
> > > Perplexed in Peoria <jimmenegay@sbcglobal.net> wrote or quoted:
>
> > > > Clearly, to have any hope at all of preserving the original
information, it
> > > > must be the case that any one tape in the population will sire on
average
> > > > more than one error-free descendent.  To complete the analogy, we need
a
> > > > selection process that is biased toward chosing error-free descendents
as
> > > > the sources for the next generation of copies.
> > >
> > > Yes, but a way of increasing the variance of the number of mutations in
> > > individuals would help selection do its job in this area.
> > >
> > > If the deleterious mutations are concentrated in some individuals,
> > > fewer members of the population need to fail for the gene pool to
> > > remain in good working order.
> >
> > I fully agree that IF recombination increases the variance in number of
> > mutations per individual, then your argument works.  What I don't
> > agree with is that recombination *will* increase the variance, unless you
> > make some totally artificial assumptions regarding the source of
> > mutations.  For example, if your mutation process inserts *exactly* one
> > mutation per individual (low variance), then recombination obviously
> > will redistribute the mutations so that you have (in the simplest model)
> > 25% with no mutations, 50% with one, and 25% with two mutations.  The
> > variance per individual has increased.  But the initial assumption
> > regarding the statistics of the mutations is unrealistic, IMO.
>
> Fortunately, your query can easily be dealt with by a more sophisticated
> model.
>
> Here's a model where mutations occur with much greater biological
> realism - where mutations occur in individuals with some fixed
> probability - and individuals can have 0, 1 or more mutations
> per generation.

Thanks, Tim, for providing the source.

Java is not my "native language", but I'm pretty sure you are wrong
about this.  Your model mutates each individual at most once per
"generation".  (The term "generation" deserves the scare quotes in
this model - an individual in this model lives for an average of 20
"generations" and is active reproductively from the moment after birth
until the moment of death).  Here is the code that shows this (one
mutation max per generation).

**** Code excerpt ****
>   private void mutateAll() {
>     for (int idx = this.mutation_attempts_per_generation; --idx >= 0;) {
>       simulateCosmicRays(idx);
>     }
>   }
>
>   private void simulateCosmicRays(int idx) {
>     if (this.random.nextFloat() < this.mutation_rate) {
>       mutate(idx);
>     }
>   }
>
>   private void mutate(int idx) {
>     int bit_mask = 1 << (this.random.nextInt() & 31);
>     this.population[idx] ^= bit_mask;
>   }
**** end code excerpt ****

Note that "idx" ranges over individuals due to the fact that
mutation_attempts_per_generation is set equal to pop_size.  If it were
set lower, only part of the population would be susceptible to
mutation.  If it were set higher, you would have an array bounds
violation.

In your model, individuals accumulate mutations throughout their
lifetime, and as soon as they accumulate enough mutations they die.
This is true for both sexuals and asexuals, but sexuals have the
chance of being born with too many mutations, whereas asexuals do
not.  A truly bizarre model.  Especially since the defective sexuals
get one "free" generation of reproduction before they have to face
selection.

[I know now how JMS felt while reviewing Wilson and Lumsden]


> Heavily mutated individuals - with more than some threshold number
> of mutations - are eliminated - and new individuals take their
> place.
>
> The results (which are highly repeatable):
>
> Asexual experiment
> Deaths: 538260
> -----
> Sexual experiment
> Deaths: 422799
> -----
>
> I.e. the sexual population needs many fewer selective deaths to keep its
> gene pool in good order.
>

Not at all surprising, since dying asexuals always carry *exactly* 5
mutations out of the gene pool, whereas dying sexuals carry 5 or
more.

I would be much happier if your model inserted all mutations at the
time of birth, and if it permitted more than one new mutation to be
inserted into an asexual newborn.

> The mutation model here is pretty good - and making it even more realistic
> would not alter the results - the effect is one that really does exist.
>
> What /could/ be criticised is the idea that there is a mutation threshold
> - beyond which you are dead.  Making this more realistic would
> reduce the magnitude of the effect - but would not eliminate it.

I agree that if you have a mutation count threshold, and the simulation
cycle runs in the order [select -> recombine -> mutate -> select again],
then you will have the effect you desire.  The reason is that when
recombination operates on the post-selection population, it is operating
on a population whose variance in mutations per individual has just been
artificially lowered by the truncation selection.  It can't help but to
increase the variance.  The effect is much smaller if the simulation
cycle is [select -> mutate -> recombine -> select again].  And the effect
disappears entirely if the selection threshold is reduced to a more
realistic number of 1 mutation per individual.  (Recall that this is a
haploid population, so "recessive lethals" are expressed.)

[Source code snipped, except for what was excerpted above.]
---
ū RIMEGate(tm)/RGXPost V1.14 at BBSWORLD * Info@bbsworld.com

---
 * RIMEGate(tm)V10.2á˙* RelayNet(tm) NNTP Gateway * MoonDog BBS
 * RgateImp.MoonDog.BBS at 1/1/05 6:01:03 AM
 * Origin: MoonDog BBS, Brooklyn,NY, 718 692-2498, 1:278/230 (1:278/230)