Re: Of Mice and Straw Men


RobinGoodfellow wrote:
> Seanpit wrote:
> > RobinGoodfellow wrote:
> >
> > > But have you actually done the calculations as described in your model?
> > > Because when I applied the calculations as per your proposed
> > > methodology, I arrived at the none-too-surprising conclusion that
> > > tempalte matching is unevolvable.
> >
> > Template matching is evolvable - not un-evolvable.
>
> Not if we are to believe your own calculations. One could easily show
> that template matching is unevolvable applying the methods you apply to
> Cytochrome C, for example.

Cytochrome C cannot be evolved using template matching. It is an
independently acting function that is not based on the pre-existence or
degree of binding with anything else. The same is true of flagellar
motility and lactase and nylonase functions. The evolution of any of
these types of functions would not be template based. Functions like
improved phage infectivity by a change in the D2 replacement sequence
or improved immunoglobulin matching to a particular antigen are
examples of template-based evolution where improved binding to another
pre-established sequence result in increased selectability. See the
difference?

> > The problems for evolution come when the functions cannot be evolved using a
> > pre-formed template.
>
> Too bad your model can't distinguish between these two cases.

They easily distinguishable.

> > Considering your position that no significant neutral gaps exist, it
> > should be quite easy to experimentally demonstrate the evolution of at
> > least one or two of the proposed steps in the pathway of a highly
> > complex system of function like flagellar motility. Why should a
> > single step should take millions, thousands, hundred, or even dozens of
> > generations to evolve if no significant neutral gaps exist?
>
> What exactly do you mean by "a single step"?

There are lots of stories out there about how the flagellum supposedly
evolved one step at a time. Take Matzke's model, for example. He
presents quite a number of proposed evolutionary steps to flagellar
formation. Well, why doesn't someone just set up one of these steps
and see if the next step in the series will actually evolve given the
proper selective environment?

> I assume a point mutation
> that grants a selective advantage doesn't qualify.

Sure it does. The problem is that none of the steps in Matzke's model
are just one point mutation away from the next steppingstone. Not at
all. They are at least dozens character differences away. Matzke, and
most other scientists, don't think that a few dozen residue differences
are much of a problem given millions and billions of years. But, they
are. They are a huge problem.

> But what about
> evolution of a new single-protein function, or an additional function
> by an existing protein or group of proteins?

All examples of real evolution in action. The problem is, none of
these examples end up beyond very low levels of functional complexity.
Not one of these examples produces a function with minimum sequence
requirements that go beyond a few hundred fairly specified residues.
Quite interesting - isn't it . . .

> All these represent
> individual steps of the same degree of complexity that we would have
> probably seen in the course of flagellar evolution, had we been able to
> observe it.

Actually, they don't. The average gap sizes are smaller at such lower
levels. The density of beneficial sequences is much higher. So, the
odds of successfully evolving a beneficial protein sequence of only a
few hundred fairly specified residues is almost infinitely greater than
the odds of evolving the next step in flagellar evolution beyond a few
thousand fairly specified residues.

>(Assuming, of course, that the flagellum did evolve and
> wasn't magicked into place by an unknown number of unspecified
> designers.) Your claim is that not only could the flagellum couldn't
> evolve, but anything on the same level of complexity as the flagellum
> couldn't evolve either. So, what complexity level must be reached by
> an invidual step in order to convince you?

I've told you over and over again - you need to show non-template type
evolution beyond a few hundred fairly specified residues. I've
specifically drawn the line several times at a couple thousand residues
or even 1000aa.

>
> [snip]
>
> > > > So, you see, with higher and higher levels of functional complexity,
> > > > the average random walk distance/time really does increase
> > > > exponentially - in a parabolic-like curve.
> > >
> > > You really don't know what these words mean, do you? Any parabola can
> > > be represented by a quadratic equation. A parabolic-like curve is by
> > > definition non-exponential, unless you think quadratic and exponential
> > > functions are the same thing. (Hint: in an exponential function, the
> > > variable - e.g. sequence length - appears in the exponent, not in the
> > > base.)
> >
> > I'm talking about the shape of the curve as it starts out at low levels
> > of functional complexity. It starts out rather flat and then gradually
> > increases in a somewhat linear fashion. Then, at higher and higher
> > levels it starts to curve upwards. Then, very quickly it starts to
> > follow an exponential increase . . .
>
> Oh, dear! Sean, you are in dire need of a remedial algebra class, at
> high school level. Do you know that the behavior you describe is
> exhibited by a very broad class of functions, including the very
> parabolas you mention, as very as every polynomial function in
> existance?

Exactly - so how is my statement mistaken?

> (e.g. L^3, L^4, e.t.c.) The shape of the curve itself
> tells us virtually nothing.

I never said it did. It was merely a statement to provide a visual.

> What matters is whether the data fit the
> exponential predicted by your model,

That's right . . .

> C^L / N(L), where N(L) is the
> number of strings of length L, and in Zach's simulation, can be upper
> bounded by Z - L, where Z is the total text length (i.e. the number of
> charactes in Hamlet).

It cannot be upper bounded by Z-L at higher and higher levels of
complexity because of the problem of a more and more limited library of
options with increasing minimum sequence size and/or specificity
requirements.

> Your prediction miserably fails to agree with
> the actual data, and in fact the disagreement actually becomes worse
> and worse as L increases.

Keep going. Zach's programs haven't gone beyond the very lowest rungs
of the ladder. The statistics do indeed start to approach C^L/N at
higher levels of functional complexity. They never actually reach
C^L/N, but they do start to approach it. They do end up producing an
exponential curve.

> > > > This is a prediction of my model. Ironically, Zach's programs actually support this
> > > > prediction.
> > >
> > > You have got to be kidding! Download Zach's phrasenation summary,
> > > http://www.zachriel.com/Phrasenation/Phrasenation.zip, and examine the
> > > file "regression.jpg". The curve of length vs. log(# of mutations) is
> > > clearly sub-linear: meaning the relationship between length and number
> > > of mutations is sub-exponential.
> >
> > Of course the rate of change is initially sub-exponential. That is
> > because of the lava-lamp process. Vertical connections stay more
> > closely connected for a while at lower levels. However, as you keep
> > moving up the ladder, these lava lamp columns start to break apart and
> > the little blobs of beneficial sequences break away, in an exponential
> > manner, from all surrounding beneficial sequences - including the
> > underlying less complex sequences.
>
> OK, then, give me one example of a long English string that is broken
> apart from other strings in this "lava-lamp-like" fashion of yours.
> Seriously, just one example will do, where you believe it would take me
> an exponential amount of time to go from a long phrase to a longer or
> shorter phrase. Surely if this "exponentially increasing gap" idea of
> yours are true, you can think of some English text that can serve as an
> appropriate illustrative example for your claims.

Take any English paragraph that describes how to do a particular task
in the shortest number of characters. If this number is a minimum of
at least 1000 characters in length, you will not be able to evolve that
sequence using evolutionary mechanisms of random mutation (of any type)
and function based selection where each step is meaningfully beneficial
in a given environment. You can even use one of the paragraphs in this
discussion if you want. Remember though, you have to use random
selection from a limited pool of options. The pool can be very large,
but it has to have an upper limit on size and it has to start at a
lower level.

> > > Please, Sean, at least try to learn what the terms mean before using
> > > them. You keep going on about "exponentially expanding neutral gaps",
> > > and yet I have a strong suspicion that you don't have a clear
> > > understanding of what "exponential" means. And then you wonder why
> > > people have trouble understanding what you're saying.
> >
> > Exponential = C^N so that for every increase in C, the random walk time
> > increases by a factor of N. If C = 20 and N = minimum sequence length,
> > then, for a fully specified sequence of 5, the average random walk
> > distance would be 20^5 or 3.2 million steps/mutations. Of course, at
> > lower levels of complexity this calculation doesn't hold true because
> > of the way in which the beneficial sequences are interconnected - not
> > completely randomly arranged in sequence space.
>
> So, you are saying that at higher levels, the "beneficial sequences"
> are "completely randomly arranged", or at least more "randomly
> arranged" than at lower levels?

Yes. They do indeed become more and more separated at higher and
higher levels. They are no longer as tightly clustered and
interconnected as they were at lower levels.

> How do you propose we measure the
> degree to which they are "randomly arranged"?

Minimum differences in residue or bp sequences.

> At what level of
> complexity might we start seeing the level of "random arrangedness"
> that agrees with your calculations?

Greater than a few hundred fairly specified residues at minimum.

> Why is it conveniently located at
> a point where experimental data is impossible to obtain?

Why is experimental data impossible to obtain beyond a few hundred
fairly specified residues if the neutral gaps really aren't a problem?
Why do real experiments with evolution start to stall out, in an
exponential manner for functions that go beyond a few hundred fairly
specified residues at minimum?

>And finally,
> how can you write, presumably with a straight face, that Zach's data
> bear out your predictions because they drastically disagree with them
> at the only "level of complexity" where they are actually testable?

Because, my predictions actually assumed from the very beginning,
before Zach came along, that low levels were indeed evolvable and
highly interconnected. I wrote out the cat to hat to bat to bit to big
to dig to dog sequence in one of the first papers I wrote on this whole
topic back in 1997. The interconnecting web is obvious at such low
levels of minimum sequence length and/or specificity. That's all that
Zach's programs prove - which is right in line with my own predictions.


> > This is what you and
> > Zach love to repeat over and over again without doing much
> > consideration beyond very low levels of complexity where this assertion
> > of yours is indeed true. For instance, Zach hasn't gone beyond
> > relatively short single word sequences in his evolutionary programs.
> > He hasn't even come close to my initial challenge of evolving, without
> > template matching, a meaningful English-language sequence requiring a
> > minimum of over 1,000 characters.
>
> Here's a deal. You provide me with a program that can determine if a
> particular set of 1,000 characters represents a meaningful English
> sentence, and I'll happily write you a program that evolves meaningful
> English strings of 1,000 characters or more. I'll even give a
> prediction of how long it might take: if I were using Zach's model of
> string evolution, I believe we would need at most 1000^4, or one
> trillion evolutionary steps in order to do so (but a probably a lot
> less).

You fail to consider the odds that a particular multicharacter
sequence, that is required to fill the gap of the next step at higher
levels, actually exists in your genomic library.

> Meaning that, if you want the results quickly enough, your
> program should be able to determine whether an English string is
> meaningful at a rate of one million per second (then, the computer
> would need around 12 days to evolve the first 1,000 character string).

It will never happen because of the problem of limited library size.

> If you have such a program, please send it to me, but don't forget to
> contact ACM first to claim your richly deserved Turing Award, because
> this would probably be the greatest single achievement in the history
> of computer science to date.

Right . . . The program is called, "Real Life". There is lots of real
time data in biology that bears out my calculations - not yours.
Nothing evolves in real life beyond a few hundred fairly specified
residues.

> > You see, at higher and higher levels
> > of complexity these interconnections that do indeed exist at low levels
> > start to break apart at a more and more rapid rate until this
> > exponential assumption of mine does indeed become more and more true at
> > higher and higher levels of functional complexity. This creates a
> > curved increase from the "sublinear" to the truly "exponential"
> > pattern.
>
> It is really a shame that you have no actual data to confirm this
> little "fact" of yours, and can't even predict at what level of
> "complexity" such a shift will start to occur. But why bother with
> such encumbrances when you can simply repeat your claims ad infinitum?

I have lots of real data to confirm this little "fact" of mine - the
real life stalling out of evolutionary processes at only a few hundred
fairly specified residues.

Sean Pitman
www.DetectingDesign.com

.