Evolution Icon Evolution
Intelligent Design Icon Intelligent Design

Yes, Winston Ewert’s Dependency Graph Is a Real Model

Image: A dependency graph, by c4ristian, CC BY-SA 4.0 , via Wikimedia Commons.

In my previous post responding to YouTuber Gutsick Gibbon, aka Erika, I linked to Winston Ewert’s 2018 BIO-Complexity article, “The Dependency Graph of Life,” as a more technical explanation of how common design can yield a dataset with some tree-like structure. Winston’s central thesis is that the nested hierarchical pattern observed in subsets of genes is better accounted for by a dependency graph which reflects the fact that programmers re-use similar coding modules in different independent systems to fulfill similar functional needs. Erika investigated this a bit, and dug up a Peaceful Science dialogue and Todd Wood’s blog. At 16:02 she says:

Basically [Winston is] proposing that the tree of life is kind of an illusion of the tree of life and that actually it is more akin to updating software which is actually quite strange.

She then goes on to say at 19:08:

But then [Winston’s model] is not really a model that’s actually distinguishable from evolution just like intelligent design is not distinguishable from evolution. You got some cross symbolism going on here.

A Distinct Pattern

But Winston’s model is distinguishable from evolution! As he wrote in his paper that first described the dependency graph model:

[The dependency graph and common ancestry models] differ on one crucial point: a species conventionally has one ancestral species, but a module typically has multiple dependencies. For example, see “echolocation” and “marine” in Figure 3. All marine species depend on a marine module, and the echolocating species depend on the echolocation module. The dependency graph is essentially a tree with extra flexibility; the modules can explain genes shared between species thought to be only distantly related by common descent. A module is not restricted to reusing code from a single source, but can freely reuse from multiple sources. Compare this to common descent where each species must almost exclusively draw from a single source: its ancestral species.

This model predicts a distinct pattern from common descent. As an example, the dependency graph predicts that modules not corresponding to taxonomic categories (i.e., re-usage of components in a manner that doesn’t fit a tree) should be abundant. In case you’d like to read the conclusion from Winston’s work:

The predictions of the dependency graph hypothesis set out in this paper have been shown to be correct. The biological data was a better fit to a dependency graph than to a tree. The data produced by a simulated process of common descent was a better fit to a tree than to a dependency graph. The data produced by a compiler was both a better fit to a dependency graph than a tree, and a better fit to a tree than to the null model. The inferred biological dependency graphs contained were not simply the tree of life with a few additions, but instead contained many additional modules.

Does Tree-Like Data Even Imply Common Ancestry?

One interesting observation by Erika, originally raised by computational biologist Joshua Swamidass at Peaceful Science, is that human diversity doesn’t always show a tree pattern. The argument is that even in a case where we all agree common ancestry is true (i.e., within humans), we don’t necessarily find a tree-like dataset. This is supposed to get common descent off the hook, so it isn’t falsified or challenged in the numerous other cases where the data isn’t tree-like. A couple of points in response:

  1. Common ancestry historically has predicted a tree-like pattern. If it doesn’t, then why does virtually every phylogenetics study evaluate the consistency index — i.e., why does it report statistics that tell us how “tree-like” a dataset is? Moreover, if common ancestry doesn’t predict some sort of a tree, then why do phylogenetic studies always prefer to report the most “tree-like” pattern for describing their dataset as the most parsimonious “tree” implied by the data? The whole premise of Baum et al. (2016) is that common ancestry is a better explanation because it seems to be more tree-like than random in its distribution. 
  1. Now it’s true that there are known processes in biology that can produce non-tree-like patterns, but it remains the case that the more tree-like a dataset is, the more confidence we have that it was produced by common ancestry. Many datasets may land in the gray zone where skeptics may feel they’re too un-tree-like to be produced by common ancestry, and believers in common ancestry feel that they’re too tree-like to be produced by common design. Many conversations remain to be had, but whatever the answer is, we need realistic assessments of what design-based models would predict. I hope this post has helped move the conversation forward in a positive manner. 

Before concluding this series tomorrow, I will finish with this for now: the fact that known natural biological processes can produce non-tree-like patterns is problematic not for ID proponents but for proponents of common ancestry. If, when we know genetic inheritance is occurring, we don’t see a tree pattern, then what does this say about our ability to test or demonstrate common ancestry? The presence of non-tree-like data, however, is not a problem for ID proponents who challenge common ancestry. That is because we feel we can account for genetic similarity better through common design.