Richard A. Watson: Compositional Evolution

Wondering if anyone here has read the book Compositional Evolution by Richard A. Watson and what your thoughts are. I am particularly interested in hearing from computational biologists.

Compositional Evolution: The Impact of Sex, Symbiosis, and Modularity on the Gradualist Framework of Evolution

1 Like

I did

1 Like

No, I’ve not read it yet, but I think I may need to bump it up my reading queue.

1 Like

The following assumptions are pervasive in our understanding of evolutionary difficulty in both natural systems and computational problem solving:

  • There must always be some path of successive slight modification conferring monotonically increasing fitness approaching any evolvable feature or adaptation.

  • There is a meaningful neighborhood metric on which a fitness landscape may be based, giving a meaningful notion of local adaptation and local hill climbing for evolutionary change.

  • Ruggedness in a fitness landscape, and the existence of local optima, corresponds to evolutionary difficulty.

  • A population will, for the most part, be more or less converged around a local fitness peak, i.e., stuck on a local fitness optimum.

  • A system where the removal of any one part causes nonfunction (irreducibility) is difficult for evolution.

  • Although epistasis is not necessarily absent, natural selection will act primarily on the additive effects of gene substitutions.

  • Large adaptive changes should be relegated to “hopeful monsters.”

All of these assumptions and notions of evolutionary change and evolutionary difficulty are based, directly or indirectly, on the assumption of gradual change.

  • Compositional Evolution, pp. 41-42

When you say “our”, who do you mean?

Not true. There must be a path, but it could include both neutral and slightly deleterious changes and still be followed. “Slight” might also be open to interpretation.

Not sure what that means, but as far as I can tellit seems to apply to our ability to create model fitness landscapes and not to actual evolutionary processes.

Seems poorly stated. The existence of local optima would certainly make finding the global optimum more difficult, but what else is implied here?

Not sure of the point to that either.

That doesn’t follow unless the parts are invariant and evolution necessarily involved addition or subtraction of those invariant parts.

Very unlikely, and very few people hold that notion.

I am not impressed by that list of bullet points. Was the list directly quoted from the book? If so, I am not impressed by the book, assuming that was a fair sample. While I’m sure that computer scientists have made contributions to biology, this does not at first glance appear to be one of them.


…this book offers: (1) a model of sexual recombination using a genetic algorithm that shows that it is algorithmically distinct from mutation hill-climbing methods; (2) a model of symbiogenesis and the major evolutionary transitions that shows that such processes are also a fundamentally different class of adaptive process from the gradualist evolutionary framework of linear incremental improvement; and (3) a framework for understanding both symbiogenesis and sexual recombination as instances of a general class of mechanisms that I have termed compositional evolution: evolutionary processes involving the combination of systems or subsystems of semi-independently preadapted genetic material. This is contrasted with the gradualist framework of evolution that depends on the linear or sequential accumulation of random modifications.

p. xiv

Yes, definitely sounds like a computer scientist’s gift to evolutionary biology. Thanks loads.


It appears that in attempt to garner interest I have instead garnered disinterest. :slight_smile:

1 Like

For what it’s worth, I’m still interested and have the book on order. Everything you’ve quoted so far strikes me as fairly typical of extended evolutionary synthesis advocates, and I already knew that particular book series drew authors from that school of thought. I’m just curious about the modeling approaches.

I’ve got my copy. Fair warning, it will take me a while to get through in full because of its technical nature.

I am at least sympathetic to the premise as laid out in the preface & introduction. It makes sense to me that combinatorics could search differently than linear exploration. But we’ll see if Watson makes a good case on the particulars.

Incidentally, one of my catch phrases at work is “Cross products get big fast.” I am hardly the first to notice this; that honor goes back at least to the Sumerians. Instead, I am usually trying to remind everyone (myself included) that we need to be mindful of the combinations we are presenting to our users. A few options here, a few options there, and all of a sudden we’ve asked the user to try hundreds or thousands of possible combinations. I’ve done it to myself, attempting to slice data along a few different axes before I remember that I won’t be able to effectively review the 10,000 charts that would result.

I also really appreciated John Maynard Smith and Eörs Szathmáry’s The Major Transitions of Evolution, so I’m also up for any further exploration of evolution along those lines, especially if there is a way to formalize those transitions.


Yes, thank you. When shopping for Compositional Evolution, I browsed through all of the Vienna Series titles and noticed that one. Definitely of interest, but one book at a time. :slight_smile:


That’s my problem. Too many books at once.

1 Like

A couple of thoughts from the first chapter.

  1. Watson talks about decomposing problems as the way in which compositional methods solve problems in polynomial time (or faster) when hill-climbing would take exponential time. As I understand it, deep learning methods can also have a divide-and-conquer aspect. I wonder if there are similarities beyond that superficial, qualitative observation. In particular, I wonder if there is overlap of the problem spaces where the two approaches offer advantages.
  2. The symbiosis-inspired methods specifically have the advantage of combining multiple ‘modules’ into a higher level of organization. I wonder if there is a tradeoff in terms of a subsequent limitation on the kind of innovation that can occur at the lower levels of organization. I’m thinking specifically of the fact that multicellular organisms are far more restricted metabolically than bacteria are. I wonder if there is a way to generalize that concept.
1 Like

Those are some good thoughts. Thanks.

Chapter 2 is interesting. As a summary of “mere evolution” – that is, the processes of descent with modification and selection – it is fairly straightforward. As a discussion of the limitations of simple genetic algorithms (those without any crossover or recombination), it is also seems reasonable and easy enough to follow. (Although I cannot comment on how often that simple version is implemented relative to others.)

Where things get a little muddier is the connection to evolutionary biology. Sure, Darwin might not have been familiar with sexual recombination. But by the time the evolutionary mechanism and population genetics were formalized to a degree that would permit the kind of analysis Watson gets into, sexual recombination was included. So I’m not sure if there is a relevant critique of evolutionary biology here.

It is also interesting how frequently Watson cites Behe and irreducible complexity. He seems to buy Behe’s critiques as legitimate with respect to gradualism but again his gradualism is such a limited view of evolution biologically that I wonder how relevant it actually is.


It appears that what he writes might apply to life before sex. He includes sex as part of his mechanisms of compositional evolution. I guess the question is whether there is an algorithmic path from non-composiitonal evolution to compositional evolution. The evolution of evolution. :slight_smile:

Possibly, although the frequency of horizontal gene transfer in bacteria suggests otherwise.

Yes. He is clearly aware of the breadth of evolutionary biology overall. And he has made it explicitly clear that he thinks biological mechanisms overall are sufficient to explain extant biological diversity. I’m just not sure how his critique of gradual/noncompositional evolution on algorithmic terms applies to actual evolutionary natural history. In other words, it may not be a question of whether there is a path from noncompositional to compositional evolution biologically because both may have always been options from the start.

And to be fair… maybe that is a point Watson is building to. Or maybe he isn’t all that interested in natural history and just wants to differentiate classes of algorithms. If either of those is the case, then fair enough.

Another good point! I admit I’m falling behind, but the weekend looms :slight_smile:

An interesting question about HGT would be whether it involves pre-adapted sequences (compositional evolution) or just any old random stretch of DNA (darwinian gradualism).

Thank you for your comments.

Both. Although bacterial genomes have far less ‘junk’ than eukaryotic genomes, so “any old random stretch of DNA” may be the less common scenario. But mechanisms for both are definitely available. For example, phages can sometimes packages portions of bacteria DNA and transfer them, without much regard for what the DNA encodes (if anything). On the other hand, plasmids often contain preadapted sequences.

Sure; it’s an interesting book and I’m happy to chat about it. Thanks for kicking it up my reading queue.