So tell us what you have learned about ev.
What is Rfreq and Rseq?
Rfreq is a measure of the information required to locate a valid binding site in a genome.
Rseq is a measure of the information conserved in the binding site.
What is ev?
A program written to show that the naturally observed phenomonen of Rseq ~= Rfreq in real genomes is easily evolvable. The genes in the program represent proteins that can bind to DNA, thus controlling their (or other's) transcription -- similar to how known genetic control mechanisms work.
In the program, the fittest invididuals have the least mistakes, where a mistake is a) a gene that cannot control itself (its protein doesn't bind to its binding site), b) a gene whose protein binds somehwere other than its binding site.
Why can't the ev program be used for the purposes Kleinman contends it can?
-The population size in ev remains constant (every generation only half the population is killed off, and then the fittest reproduce to fill the gap back up) whereas in every single empirical example Kleinman has cited the population dies off in only a few generations.
-The three selection pressures in ev each target every single potential binding site in the genome (something Kleinman has repeatedly said to be impossible in reality, mind you), instead of individual genes / bases.
Why is Kleinman wrong with regards to the mistake count selection pressures?
There are
three global selection pressures active in ev. First, mistakes are generated for
missed sites, which is when a gene's protein doesn't bind to its own binding site. Second, there is
spurious binding within gene, which is when a gene's protein binds to its own gene but not at the control site. Third, mistakes are generated for
spurious binding outside gene, which is when a gene's protein binds to a site outside of the gene. In ev, users can assign a relative weight to how bad each of these mistake categories are. I will use vector notation, I.E. [1,1,1] meaning each has an equal weight of 1, [1,2,3] meaning missed sites have weight 1, spurious binding within gene has 2, etc.
Furthermore, there are two metrics one can use when determining when to stop the simulation. First, one can stop when a perfect creature is evolved, which is defined as an individual with no mistakes -- when all of its proteins can bind to their own binding sites. Second, one can stop when Rseq >= Rfreq, or when the control is nearly perfect. Keep in mind, however, that the
purpose of ev is to research the
latter and it was written with that in mind.
Now, Kleinman's contention is that ev supports his hypothesis because the number of generations it takes to evolve a perfect creature with weights [1,0,0] is much less than the number of generations it takes with weights [1,1,1], or in other words that the evolution under one global pressure is much faster than under three. Let us examine whether this claim has merit.
First, what is a perfect creature when the weights are [1,0,0]? By definition, it is a creature who's genes all produce proteins that bind to their own control sites. That is good. But it also has a myriad of proteins that bind in many other places (because no penalty is given to spurious bindings). This is bad. In fact, it would not happen in reality, for reasons that should be obvious. Already, we have shown that Kleinman is incorrect in this matter, but for the sake of completion, let us proceed.
Second, what about using Rseq >= Rfreq as the stopping point? It turns out that with weights [1,0,0] the simulation will
never attain it, even with extremely small population and genome size. This should be a pretty good indicator that the program was not designed to run with weights [1,0,0].
Third, what about using other combinations, like [0,1,0] or [0,0,1]? Mathematically, if Kleinman wants to treat each of the three pressures similarly, then they should in fact be similar and thus lead to similar results when they are the only active pressure. This, however, is not the case. Both [0,1,0] and [0,0,1] evolve a perfect creature in only
one generation, meaning that the garbage the genome is initialized with is in fact considered "perfect" under those weight distributions. This is quite different than [1,0,0] where, using the default parameters for instance, it takes 6 generations. Likewise, neither [0,1,0] or [0,0,1] ever stop on the Rseq >= Rfreq metric.
Thus, it is utterly ignorant to set
any of the weights to zero. Note that
none of the "peer reviewed" articles on, or even citing, ev make use of such weights.
What happens when we use ev as intended?
First, we have already gotten Kleinman to admit that in reality there is
never a single selection pressure -- rather, there are instances of a single
strong pressure against a background of many weak ones. So lets look into this by making one weight 100 times as high as the others. Using the default settings and stopping on a perfect creature, here are the generations required:
a) - [100,1,1] = 360 generations
b) - [1,100,1] = 1034 generations
c) - [1,1,100] = 889 generations
d) - [100,100,100] =
662 generations
Interesting. Applying all three strong pressures at once (d) resulted in not only a higher rate than only one pressure in two cases (b and c) but also a
faster evolution. Furthermore, even using Kleinmans prized pressure, missed bindings, the
rate under three pressures was higher (because if we applied the single pressure back to back it would take 360 x 3 = 1380 generations).
ETA: similar results follow if you use the Rseq >= Rfreq stopping point.
FEEL FREE TO RESPOND TO ANY OF THIS AND EXPLAIN WHY I AM WRONG, KLEINMAN
It’s up to you to prove your code is a valid computation. You claim it shows that the greater the number of selection conditions the faster the system evolves.
Kleinman, it is code -- it is its own proof. All you have to do is look at the code.
Or are you saying that you don't know programming, and need help understanding the code? If that is the case, I would be happy to explain it, but.... well, then why do you insist you know the internals of ev if you can't even read code?
Dr Schneider’s peer reviewed and published model shows the exact opposite of what you claim and the empirical evidence substantiates it.
No, it does not.