Robert Lücking

Botanischer Garten und Botanisches Museum Berlin

I need to run GMYC but seem to be hitting a wall. The GMYC web server is not responding for hours, although the tree is quite small. I have previously also done this in R and wanted to do this now, but apparently the "splits" package which contains the GMYC algorithm does not exist anymore? There only is a package "split" which has nothing to do with "splits". Any advice would be highly appreciated.

Robert

I generally use GTR+I+G for phylogenetic analysis, as this is the universal substitution model, with other models being simplifications by equalizing rates. Model testing generally uses the argument that GTR may be overfitting, but I have not yet seen a study where that would really cause substantial effects. Usually, when I do model testing, GTR+I+G is always close to the top and the differences in likelihood and AICc or BIC are usually minor. Also, commonly I get e.g. TVM+I+G as best fitting model, which differs from GTR+I+G in that AG and GC are equal (sometimes TPM, with three equal rates). Given that model testing depends on alignment details (e.g. removal of ambiguously aligned regions changes original base frequencies), among other factors, and there are parameters other than base and substitution frequencies that may influence particular substitutions (e.g. secondary structure), I wonder whether it is actually justified to chose an only slightly deviating model over GTR, when the latter is close to the best model in AICc or BIC and actual topologies or support, when using e.g. TVM or GTR, do not really differ, except maybe for non-supported branches. Isn't it OK to use GTR per default, making no assumptions about the data, compared to fitting a marginally better model such as TVM that depends on certain details and in the end makes no or very little difference?

I would like to start a small discussion on the problem of the taxonomic paradoxon, in popular saying also known as the chicken and egg problem. This recent paper claims to solve the paradoxon, but I believe this is not the case. While one can certainly argue that even the "species" is an artificial construct, as long as we do taxonomy and we use taxonomy as a tool to address evolutionary processes, we should be able to resolve the taxonomic paradoxon of species delimitation through time. Horizontally it can be done by using parameters such as phylogeny (topology, support, branch length) and reproductive biology. However, these measures collapse entirely if we look at a lineage backwards in time. By default, we postulate that any generation stems from a prior generation belonging to the same species. Even if mutations occur and are inherited, this does per se not "generate" new species, as speciation is a lengthy process of genetic variation sorted out by environmental factors. The only instance one could imagine spontaneous speciation is where partner recognition is highly specific based on narrow chemical stimuli (key-lock principle), but that would apply to particular organisms only. So the paradoxon lies in the notion that if you compare individuals along the same lineage using a given distance, they would resolve as distinct species, but you could freely shift that intervall along the time axis and hence there would be no fixed moment of speciation, which means you would be essentially unable to assign individuals to distinct taxa along that lineage. By extension, if you had access to all living individuals ever existing on this planet, you could not classify them naturally, since even at a branching point, two subsequent generations could not be distinguished. So how to we solve this? Or are we just happy that our taxonomic classifications rely on the fact that all past individuals that could mess with them are extinct? One of the practical applications of this important question is paraphyly, and the many different opinions in this respect seem to indicate that this question can indeed not be resolved?