Evolution
Intelligent Design
Response to a Critic: But What About Undirected Graphs?
Photo: A zebra finch, Dundee Wildlife Park, Murray Bridge, South Australia, by Peripitus [GFDL, CC-BY-SA-3.0 or CC BY-SA 2.5 ], from Wikimedia Commons.
Winston Ewert, of Biologic Institute and the Evolutionary Informatics Lab, recently published an extraordinary paper. In it, he argues that the pattern of life is better explained by a dependency graph (a software engineering concept) than by Darwin’s tree of life. For previous discussion, see here, here, and here.
Critiques Roll In
Now the critiques are beginning to roll in. The paper in the journal BIO-Complexity has attracted interest and critique from a number of serious mainstream scientists.
One commentator argued that we (the scientific community) already know that life is not well explained by a tree, and says that the leading evolutionary explanation is really a reticulated tree or an undirected graph and therefore it is no wonder the dependency graph model beats a “strawman” model. He further argues that human genetic data fits an undirected graph better than a tree, and so would a dependency graph (or so he predicts). Essentially, an undirected graph is what you get when you allow for species hybridizing as well as species splitting: genetic material merges from more than one branch of the tree. This is kind of like lateral gene transfer, but more extreme.
Image: Reticulate evolution, via Wikimedia Commons.
The One Truth?
This is a valid kind of argument. Ewert was very careful to note that his method does not prove the dependency graph is the One Truth. Instead, the method demonstrates that it is a more powerful model than a tree. He then invited evolutionists to provide a better model.
It is possible that the defenders of common descent will devise a testable model that can explain the successful predictions of the dependency graph hypothesis. Perhaps some sort of extended synthesis of common descent with additional mechanisms could be that model.
So is the undirected graph a better model than the dependency graph? We should test it! However, there is another important consideration to bear in mind. What kind of undirected graph are we talking about?
The critic points out that human genetic data is well explained by an undirected graph. This is because there has been continual recombination and hybridization between human lineages throughout history. The concern is that Ewert’s method would misidentify this undirected graph as a dependency graph, implying that false positives are generally possible. But this is not a surprise: If the truth is A, but you only compare B vs. C, then of course you will get a false positive: B or C. What would be more worrying is if you compare A vs. C and it selects C. It would be interesting to check that too, and that might lead to refinements in the process.
Why Ewert Chose Metazoa
But is this concern relevant to Ewert’s results? The thing is, Ewert specifically chose Metazoan species because “horizontal gene transfer is held to be rare amongst this clade.” Likewise, in Metazoa, hybridization is generally restricted to the lower taxonomic groupings such as species and genera — the twigs and leaves of the tree of life. In a realistic evolutionary model for Metazoa, we can expect to get lots of “reticulation” at lower twigs and branches, but the main trunk and branches ought to have a pretty clear tree-like form. In other words, a realistic undirected graph of Metazoa should look mostly like a regular tree.
Photo: Zebrafish, by Oregon State University [CC BY-SA 2.0 ], via Wikimedia Commons.
For example, a realistic undirected-graph evolutionary model could easily accept hybridization between modern humans and Neanderthals, but it would have to heavily penalize anything that looks like a hybridization between, say, finches and fish.
Of Zebra Finches and Zebra Fish
And yet, as Ewert notes in his paper:
[D]espite the similarity in names, Taeniopygia guttata (zebra finch) and Danio rerio (zebra fish) are only distantly related because one is a bird and the other a fish. As such, it should be relatively improbable to find genes shared only between these two species. But according to the Hogenom dataset, there are nineteen gene families found only in this pair of species. The dependency graph model can assign high probabilities to both of these combinations by postulating a module shared between the pairs of species.
This is something that a realistic evolutionary process would be highly unlikely to do.
It will be very interesting to further explore the relevance of other models, including undirected graphs, but let us remember to keep realism in mind, too. A dependency graph is not any old ad hoc hypothesis. It was posited because it is something that we observe in software engineering. Evolution is restricted in different ways than intelligent design is.