How Does Tokuriki 2009 Affect Conclusions from Axe 2004?

[colewd (Bill Cole)]

@swamidass

@colewd, if this is the standard, it is an example of the marksman’s fallacy.

Hayashi 2006

The question remains regarding how large a population is required to reach the fitness of the wild-type phage. The relative fitness of the wild-type phage, or rather the native D2 domain, is almost equivalent to the global peak of the fitness landscape. By extrapolation, we estimated that adaptive walking requires a library size of 10^70 with 35 substitutions to reach comparable fitness. Such a huge search is impractical and implies that evolution of the wild-type phage must have involved not only random substitutions but also other mechanisms, such as homologous recombination. Recombination among neutral or surviving entities may suppress negative mutations and thus escape from mutation-selection-drift balance. Although the importance of recombination or DNA shuffling has been suggested [30], we did not include such mechanisms for the sake of simplicity. However, the obtained landscape structure is unaffected by the involvement of recombination mutation although it may affect the speed of search in the sequence space.

How do you explain the existence of the wild type given the library size required to find it?

[Agauger (Ann Gauger)]

@colewd

Have you read the Hayashi paper?

@colewd @Rumraket
Can one of you give me the citation for Hayashi? I seem to have dropped in on a conversation in the middle.

[Mercer (John Mercer)]

@colewd

This is not my argument. My argument is that all the papers are relevant but there is a range of results based the functional requirements.

You are employing the sales technique of anchoring, possibly unwittingly. You clearly are not approaching this with an open mind.

Dr. Gauger has set the functional requirements clearly:

@Agauger

But even “radical changes” have to go from no function to weak function, since random mutation is highly unlikely to achieve even moderate levels of function right off the bat. So the transition is of interest.

@colewd

Have you read the Hayashi paper?

Yes, and others. How many papers have you read?

[swamidass (S. Joshua Swamidass)]

@colewd

How do you explain the existence of the wild type given the library size required to find it?

Do you believe God intentionally and directly created/designed HIV, malaria, Herpes, and/or Hepatitis viruses?

[colewd (Bill Cole)]

Here it is:https://dx.doi.org/10.1371%2Fjournal.pone.0000096

[Mung]

@swamidass

Do you believe God intentionally and directly created/designed HIV, malaria, Herpes, and/or Hepatitis viruses?

I do.

Do you believe that God did not intentionally and directly created/designed HIV, malaria, Herpes, and/or Hepatitis viruses?

Please present your evidence and state your reasoning.

[Mercer (John Mercer)]

@Agauger

From what I’ve read, the method can screen a library of 10^10 in each round, with successive amplifications and panning for binding at each round. That was from an older review. Maybe it’s up to 10^12.

Why would you cite the number screened but ignore the frequency of hits?

[swamidass (S. Joshua Swamidass)]

@Mercer

From what I’ve read, the method can screen a library of 10^10 in each round, with successive amplifications and panning for binding at each round. That was from an older review. Maybe it’s up to 10^12.

Why would you cite the number screened but ignore the frequency of hits?

That is relevant. It marks out the limits of what we could expect to find in the lab. The fact that we can’t screen more than 10^12 means that any function we determine doesn’t need more than this amount of probabilistic resources to find. Of course, it could be even less required, but it would not require more. This is important, because we have found an awful lot of useful stuff with just 10^12 (or less) searching capacity, even though evolution has well over 10^43. This also means that anything we find in the lab is far more likely than what Axe would predict, 10^77.

[Mercer (John Mercer)]

@swamidass

That is relevant. It marks out the limits of what we could expect to find in the lab.

Only if we know the frequency of hits.

@swamidass

The fact that we can’t screen more than 10^12 means that any function we determine doesn’t need more than this amount of probabilistic resources to find. Of course, it could be even less required, but it would not require more.

Yes, but the fact that many catalytic antibodies have been found by immunizing mice shows that it’s much easier, as the immune repertoire is a hundred-fold smaller, at 10^8.

@swamidass

This also means that anything we find in the lab is far more likely than what Axe would predict, 10^77.

Yes, Axe’s prediction is off by an incredibly large factor, as Art’s illustration shows. I think that “far more likely” is far too modest as a descriptor. “Incomprehensively more likely” is more accurate.

[Mercer (John Mercer)]

@colewd

How do you explain the existence of the wild type given the library size required to find it?

Heritable variation and selection, probably with a good dollop of drift.

@Agauger

But even “radical changes” have to go from no function to weak function, since random mutation is highly unlikely to achieve even moderate levels of function right off the bat. So the transition is of interest.

[Agauger (Ann Gauger)]

@swamidass

That is relevant. It marks out the limits of what we could expect to find in the lab. The fact that we can’t screen more than 10^12 means that any function we determine doesn’t need more than this amount of probabilistic resources to find. Of course, it could be even less required, but it would not require more. This is important, because we have found an awful lot of useful stuff with just 10^12 (or less) searching capacity, even though evolution has well over 10^43. This also means that anything we find in the lab is far more likely than what Axe would predict, 10^77.

Just a little context is necessary here. If you are going to make a case against Axe you need to address what he was claiming. Axe was not asking about functional shifts, or binding to columns. These have been accomplished in the lab, within limits, and they do not involve a change in the protein’s fold. He is talking about the rarity of functional protein folds. Folds are the highest level of category for protein structure. To go from one fold to another requires a change in topology. I know of only one paper that reported a switch in fold based upon human engineering in the lab. And brilliant engineering it was. Don’t let them fool you
you into thinking a single base change did it. Read what it took to get the two proteins ready for the switch. http://www.pnas.org/content/pnas/106/50/21149.full.pdf

There are bi-functional fold switching proteins found in nature, but that would seem to require a special design also.
http://www.pnas.org/content/pnas/early/2018/05/16/1800168115.full.pdf
The title is a bit of an exaggeration.

[Rumraket (Mikkel R.)]

@colewd

How do you explain the existence of the wild type given the library size required to find it?

Among numerous entirely plausible explanations, the wild-type D2 domain occupies a different peak in the fitness landscape and evolved from a different location in sequence space than the starting protein used in the Hayashi experiment, which just climbed up the nearest hill and got stuck there.

How do you explain that the random protein was just by chance able to perform the function of the D2 domain of g3p if functions are supposed to be as rare as 1 in 10^77 sequences?

[Rumraket (Mikkel R.)]

@Agauger

He is talking about the rarity of functional protein folds . Folds are the highest level of category for protein structure. To go from one fold to another requires a change in topology.

I’m sorry but you appear to be contradicting yourself, because earlier in this thread you wrote:

Doug did not want to measure where buffering was taking place, neither did he want to measure during exponential decline. He wanted to measure at the threshold of enzyme activity, because he was trying to quantify the number of changes to go from no function to function. This is the transition boundary that is of interest to evolutionary biologists.

So which one is it? Was Douglas Axe trying to see how difficult it was to turn one functional protein into another with a completely different fold, or was he trying to see how difficult it is to turn a nonfunctional sequence into a functional one?

You say that Axe was not “asking about functional shifts, or binding to columns”. But a protein binding to a substrate it previously weren’t capable of, is an example of going from no function to a function, which is one of those transition boundaries of interest to evolutionary biologists.

[Agauger (Ann Gauger)]

@Rumraket

So which one is it? Was Douglas Axe trying to see how difficult it was to turn one functional protein into another with a completely different fold, or was he trying to see how difficult it is to turn a nonfunctional sequence into a functional one?

@Rumraket
He is not trying to turn a protein’s function into another in the 2004 paper. And that’s not what I said, precisely.

@Rumraket

He is talking about the rarity of functional protein folds .

He wanted to know how difficult it is to go from non-functional to functional for a specific functional fold. He sets the threshold low for a particular reason.

There are two approaches to asking about the rarity of functional folds. The first, the forward approach, starts with some sort of random sequence, then selects for improvement in some basic function, such as binding to a molecule like ATP or some other hapten. The second is to start with a functional fold, and establish the threshold at which that fold goes from functional to non-functional, and then estimate from that what number of sequences are capable of carrying out that function.

Let me upload an image from his paper. This has to do with why he chose to use a low threshold and weakened enzyme, and why his numbers differ from forward approach studies, where you go from random sequence to some binding function or weak catalysis.

In discussing this figure he says:

How might this picture be reconciled with the much higher prevalence of function often reported
in studies using the forward approach?_ Figure 9 illustrates two possible ways for functional
sequences to appear relatively common when a very low functional threshold is used. Figure 9 (a) represents a global-ascent model of the function landscape, meaning that incremental improvement
of an arbitrary starting sequence will lead to a globally optimal final sequence with reasonably high probability. In this case, sequences exhibiting function at any level are properly regarded as
suboptimal versions of the optimal archetype. Consequently, if we want to know how common
sequences of this functional type are (regardless of optimality), we should set the functional threshold
as low as possible. The higher of the two thresholds shown in Figure 9 (a) would therefore lead to a
considerable underestimate. However, if the real landscape is more like the local-ascent model depicted in Figure 9 (b), where incremental improvement leads to an archetypal sequence for only a relatively tiny set of local starting sequences, then the lower threshold would lead to a considerable overestimate. In essence, activity might be a reliable marker of archetype-like mechanism down to some minimum level, but not below.

If there are lots of ways to achieve local optima related to a particular function then they will appear to be numerous. But they won’t be able to be improved much. If there is a smooth global optimum, There will be fewer candidates at the beginning, but it should be relatively easy to optimize once the first level is achieved.

People keep getting the 2004 paper, which Art Hunt critiqued and which has to do with the rarity of functional folds in sequence space, with my paper with dDoug, where we looked at the conversion of one protein’s function into another protein’s with the same fold. Different scale questions.