Comments on Gpuccio: Functional Information Methodology

Yes… do you dispute that?
Given X trait, and Y changes in environment, do you expect more than 50 percent to lead to better optimisation of X.
I am assuming that a change in environment which optimises a preselected trait will be in the minority.

The genes coding for functioning antibodies are not there in any human being. The function of antibodies is to bind to pathogens. No part of the DNA you were born with coded for antibodies that match more than a negligible fraction of pathogens. The only way that genes for functional antibodies are produced is through multiple rounds of random mutation, first through VDJ recombination and then through somatic hypermutation.


Dispute it? I don’t know, I need to see evidence to be convinced one way or the other. It’s not all that clear what you’re saying. Which was basically my point to begin with. We need specifics, and we need numbers.

For example, “more than 50 percent to lead to optimisation of X”
More than 50% of what? Species?

Talking about a trait X and environmental change Y, what are you measuring exactly? I’d expect us to measure mean fitness, and we’d have to specify the enviromental parameter undergoing change. What is that? Mean yearly temperature? Humidity? Wind? Etc.

1 Like

I’ll just go get some :popcorn:

1 Like

i actually refer to different sequence that code for the same function. even according to evolutionery sources (like talkorigin) the chance to get the same function again is about 10^40 for a 100 aa long protein (cytochrome c for instance). so it’s very unlikely to get the same function again.

Look at this. What this means is that Gpuccio will look at some particular protein sequence, and if we don’t know of other protein sequences that can perform the same function, he will define the FI to be high because the target space is small.
So he will say a protein with this function could not have evolved, because the target space is too small to have been found by a sort of “blind search”.

If we then show that in fact the protein could have evolved from other proteins with different functions, or perhaps that there are many other proteins that can perform the same function, Gpuccio will then say the target space is huge, but that means the FI is low and so the function evolving is trivial.

So Gpuccio has defined his 500-bits rule such that if something has 500 bits, it can’t evolve, and then says it’s our job to prove him wrong. If we then do that, he says we’ve just proven it doesn’t have 500 bits.

But then what use is it to bother with this whole FI nonsense? Why not just say the target space is either big or small, and this is what has implications for the probability of evolving the function. The whole deal with FI is a sideshow. It looks like it’s invented just to make it sound fancy and technical, but adds nothing to our understanding of the proteins, or the function, or the evolvability of it.

And more importantly, how does Gpuccio know the size of the target space? If you look at how Gpuccio derives the FI for some protein, he’s looking at sequence conservation for the protein in the known diversity of life. Gpuccio is taking the number of sequences currently used in extant living organisms as a proxy for how many possible proteins there are with a similar function. Obviously that doesn’t tell us anything at all about whether there are other possible proteins with a similar function

Every time Gpuccio declares that some protein exhibits 500 bits of FI because of how many similar protein sequences with the same function he can find in some public database, he’s essentially claiming to know that there are no, or not enough, other alternative protein sequences that can perform the same function, and so the target space is too small to have been found by blindly sampling sequence space from an arbitrarily picked position. Should we happen to find evidence that there ARE such other functions out there, Gpuccio will simply include those in his FI calculation which will then reduce the amout of FI exhibited by the sequence.

The whole 500-bits thing is a question-begging assertion. The whole concept has hidden the claim that evolution could not have produced X away behind a smokescreen of fancy-sounding, but ultimately vacuous technobabble.


isnt that process just improve affinity that already exist in these antibodies?

All proteins have some small mutual affinity. You’ve just proven that no new function needs to have evolved since the origin of life, and the total functional diversity of all molecules used in extant life are mere modifications of function having been adjusted up or down as needed. We can tell Michael Behe to contact you when he says evolution needs to find new functions, not merely modify or degrade (devolve) existing ones, as you’ve just invented an excuse that says he’s wrong.

But was he using the stars to find his heading?

1 Like

That’s why the function matters here. The function of antibodies is to resist infection. If we only produced antibodies from the DNA we inherited, they would not be functional.

1 Like

No, the exact same sequence again. Not the same function.

is about 10^40 for a 100 aa long protein (cytochrome c for instance). so it’s very unlikely to get the same function again.

Same sequence you mean, and randomly without selection, yes that is very unlikely.

Also, the probability of randomly generating the same 100 aa protein sequence twice is 1 in (20^100)^2 = ~1.6x10^260.

But protein sequences aren’t generated by random assembly of amino acids. When two protein sequences that share common descent evolve a few amino acids convergently, the entire sequence isn’t being randomly assembled twice. Rather, the DNA which encodes the exact same protein in both lineages, acquire a few mutations that code for the same amino acid.

So if some ancestral species has this DNA sequence: AAAAATTAAATA
Then if two of it’s descendants inherit it, and these two descendants go on to establish their own lines of descent, then eventually in one of those lineages, the gene mutates to(say):
And the mutation is favored by natural selection, so it sticks around for subsequent generations to inherit. Many other mutations happened, but they were not favored by natural selection.

And in the other lineage, the same mutation also happens:
Same thing, in this species the mutation was also favored by a similar selective pressure. Many other mutations happened in a very large population, but they also weren’t favored by natural selection.

That’s how you get convergence in sequence. Not the same thing as randomly generating the exact same sequence twice without feedback from selection.

1 Like

My comment was in response to @Roy and @Timothy_Horton claiming that variation in environment is what enables evolutionary process to innovate and that this is not reflected in evolutionary algorithms.
I would love to see some numbers too.
Or is that numbers only required from those who are skeptical of evolution and claims supporting evolution don’t need evidence?

They’ve said that. They haven’t shown it, mainly because it isn’t true - EAs can and have been used to find solutions to problems where the solution was not previously known, and so could not possibly have been smuggled it.

Dave Thomas’s Steiner tree challenge is one such case - the optimal solution to that problem was unknown to Dave Thomas until the GA was run.

P.S. Any objection that this case involves only a small amount of FI can be trivially fixed by modifying Dave Thomas’s GA to find a Steiner tree connecting a much larger number of points.


are you saying that any given protein can bind to any given molecule? it’s not what the experiments showing us. for instance only one in any 10^12 proteins can bind to ATP.

actually they are talking about the same function. they give the cytochrome c example. according to their source about one in every 10^93 sequences will perform the cyrochrome c function. thus the chance to get cyrochrome c again is one to 10^37.

indeed in the case you mentioned. but not in the case i refer to. for instance: electric organ supposedly evolved 6 times. so even if we assume that we need only about 100 aa to that change the chance for that suppose to be about 10^40 per case.

I give up.



As noted in the other thread, this is only true for EAs which have unchanging fitness landscapes, which necessarily includes non-interacting ‘organisms’.

Neither condition applies to biological evolution.

Are you saying that biological evolution cannot be defined as an algorithm?
If so, how is it a theory?

Good Lord.

I think I’m with you, @Rumraket.


No. I’m saying that biological evolution cannot be modelled using an EA which has an unchanging fitness landscape, and non-interacting ‘organisms’.

Why haven’t you tried running the EA I provided yet?