Durston's FI

[T_aquaticus]

For this reason, I have focussed, and continue to focus on protein families that have had the benefit of sufficient sampling produced by thousands of independently evolving populations across a wide range of taxa (preferably across many phyla).

From my understanding, this can only tell us which residues we can change in an initial protein without losing the function it started with. It can’t tell us how many protein sequences are available at the very start. Evolution is heavily influenced by historical contingencies (i.e. Butterfly effect), and I think this is biasing your analysis. You would also need to factor in contingencies caused by epistatic effects where neutral mutations can later become essential in stabilizing otherwise deleterious mutations.

Fitness landscape modelling would need to be a part of your analysis. Once you start with a function you are throwing away a much larger landscape of function and focusing in on just the mutations that will move up the landscape from one single starting point. I don’t see a way for your method to detect all of the possible starting points in that fitness landscape.

[glipsnort (Steve Schaffner)]

From my understanding, this can only tell us which residues we can change in an initial protein without losing the function it started with.

It tells us less than that, even. Conservation can tell us which residues we can change without reducing the functionality of the protein. There might be many mutations possible that would preserve some function – enough to provide a path to the protein – that are no longer available for evolution to take because they have lower fitness than the well-established protein that we’re examining. Adaptive evolution involves a pretty effective one-way filter on fitness changes. This means that sequence conservation is a badly biased estimator of the density of functional states near an existing, well-adapted gene.

[Mercer (John Mercer)]

For this reason, I have focussed, and continue to focus on protein families that have had the benefit of sufficient sampling produced by thousands of independently evolving populations across a wide range of taxa (preferably across many phyla).

Myosins are a perfect example of such a family:
https://www.researchgate.net/figure/Different-classes-of-myosin-superfamily-Phylogenetic-tree-of-114-myosin-sequences_fig7_221686367

Gpuccio: Functional Information Methodology

In discussions with various people in the field, there seems to be two research questions to answer (which appear to have come up here, though I’ve not read anything other than gpuccio’s one reply included in the email sent to me):

1. How can we test for sufficient sampling, given common descent and,
2. How can we test for sufficient sampling, given the possibility of a global maximum fitness in sequence space, clustering our data in only a subsection of sequence space?

There is a good test case for #2 in this family, as there are many clinical papers documenting mutations in the beta-cardiac myosin heavy chain (MYH7) gene, nearly all of which are likely to be functional, large numbers of which have been assayed for enzymatic activity, and only rarely cause disease (epistasis is the likely reason).

[swamidass (S. Joshua Swamidass)]

I don’t think there is a good case for #2. The space of sequence sampled by common descent is tiny compared to all of sequence space.

[Mercer (John Mercer)]

I don’t either. The natural, yet functional, variation we observe for MYH7 would provide a data set that likely would show that.

[Mercer (John Mercer)]

I don’t think there is a good case for #2.

You’re right. I’m saying that myosins are a good case for TESTING #2. I’ve modified my comment to clarify.