Monday, September 7, 2015

The Tree of Trees is a network

A couple of years ago this paper appeared:
Marie Fisler and Guillaume Lecointre (2013) Categorizing ideas about trees: a tree of trees. PLoS One 8: e68814.
The authors note:
We study the history of the use of trees in systematics to represent the diversity of life from 1766 to 1991. We apply to those ideas a method inspired from coding homologous parts of organisms. We discretize conceptual parts of ideas, writings and drawings about trees contained in 41 main writings; we detect shared parts among authors and code them into a 91-characters matrix and use a tree representation to show who shares what with whom. In other words, we propose a hierarchical representation of the shared ideas about trees among authors: this produces a "tree of trees."
The authors continue:
Why should we choose the tree that maximizes contiguity of identical character states (i.e. the most parsimonious tree) and not another one? [That is,] why should we choose the tree maximizing consistency among characters? ... Maximizing consistency among characters is just offering a rational interpretation of the character distribution across the compared entities, by using a hierarchy from the most general to the most particular. We prefer this hierarchical representation over networks in a first step because it is what we need to test for consistency of previous categories, propose new ones and exhibit sharings (even homoplastic ones if needed).
Unfortunately, "the parsimony analysis provides 279 trees of 378 steps, with a C.I. of 0.24 and a R.I. of 0.61". In other words, there is very little consistency among the characters; and there is very little hierarchical structure in the data, as shown by my NeighborNet analysis of the same data.

The conclusion, that "we consider that networks are not useful to represent shared ideas at the present step of the study" seems rather dubious. The tree-makers do not generally form groups, but share phylogenetic  ideas in a more haphazard manner. Nevertheless, the network neighborhoods shared by the various writers sampled do actually show quite clearly who shared tree ideas with whom.

It is interesting that the tree ideas are shared in a network manner, rather than a tree, as this indicates that there are no really clear schools of phylogenetics represented. Indeed, the writers are inter-mingled in a way that shows no development of tree ideas over time, although the various neighborhoods do tend to associate writers of similar vintage. There are no real surprises among the compositions of these neighborhoods.

Perhaps the most interesting aspect of the network, as shown, is that both Wallace and Haeckel changed their ideas about trees through time, whereas most of the other writers were more "of their time".

It is also worth noting that Buffon, Duchesne and Rühling all illustrated reticulated networks, not trees; but this is not one of the characteristics included in the dataset. The paper's authors do acknowledge that Buffon's diagram is "a tree-like extension of maps", but they fail to mention that Linnaeus also likened biological relationships to a map (not a tree), but instead treat him as part of the outgroup (see An outline history of phylogenetic trees and networks).

No comments:

Post a Comment