lunes, marzo 31, 2008

Dispersion strikes back... but in the wrong direction

Just two weeks before, a seminar about biogeography and phylogenetics starts here, in Tucumán. We were about 8 people taking about a biogeographic paper (the next week, it will be a phylogenetic one). So there are my own impressions about the papar (I posted it late, because I'm out of the town because of the of the holly week break).

Sanmartín, I., van der Mark, P., Ronquist, F. 2008. Inferring dispersal: a Bayesian approach to phylogenetic-based island biogeography, with special reference to the Canary islands. J. biogeografy 35: 428-449. DOI: 10.1111/j.1365-2699.2008.01885.x

Personally I do not like Bayesian methods in phylogenetics, they are based on several flawed principles [1], but I will try to focus in the other aspects of the proposed framework of Sanmartín et al., instead of attacking the Bayesian principles.

As many others, Sanmartín et al., attack 'vicariance' biogeography on the grounds that they only counts vicariance and ignore dispersal, I think that the tale is somewhat different, vicariance biogeographers are fully aware of dispersal, but they can't find grounds to found a general dispersal pattern, in the other hand vicariance provides a fully explicable pattern that is shared by several and unrelated organisms, then dispersal, can only be addressed group-wise. But in the case of the islands without any connection to land, vicariance can not be used as a general explanation. But can dispersal provide one?

Sanmartín et al., argue that the answer is 'yes', but instead of show why such conclusion arises they jump to their own model of dispersal, the causal mechanism that put several different organisms in the same model is never given, then their own asseveration that dispersal can produce 'concerted' (i.e. general) patterns, which is the most interesting question raised by they are never answered. What is striking is that they show several different models of dispersion proposed for canarian taxa.

Moreover the model developed, is a symmetrical one, then if in fact it is a common dispersal route, by an oceanic stream for example, it can be masked by their model that presupposes that both dispersal directions are equally probable.

Then they jump on the model frenzy, they are as embedded on the power of model approaches, that they raise 'new questions', that were irrelevant before their model. The most striking one is the 'carrying capacity' of their model. Carrying capacity is a term borrowed from ecology, and in this case is somewhat equated with island raw richness. This parameter, that seems to be originated from the 'island biogeography' of Wilson, can be of interest to ecologist, but they relevance in the search of common biogeographical patterns is never clearly showed (and I do not think that it has some empirical meaning), but now, this is an important question: most of Sanmartín et al. discussion, are about carrying capacities, not about dispersal models, which taxa are best fitted to the model, dispersal rates for different groups, or common routes of dispersal. Sanmartín et al. transform the parameter calibration into the research answer, losing the important questions in the middle.

When given the results, Sanmartín et al. contradicts many of its initial reasons to choose a likelihood model, the first one is that the likelihood of each model allows a selection of the models, but as the model (and its parameters) with the best likelihood seems to be illogical, they move on a sub-optimal models, they claim that it is not enough data, but when is enough data? How can I selecet a model only guided by its unexpected results? Why no limit the model to certain values if we are afraid of illogical answers? I think that Sanmartín et al. Results actually undermined their claims.

Sadly the paper is more like a discourse from a politician in campaign: all about the promises of wonderful results that might be open for likelihood models, but the actual use of the model with the empirical case is disappointing.

If a dispersal biogeography based on 'concerted' patterns of dispersal is to be used, their user would answer several questions: (1) why expect a 'common ' dispersal pattern? (2) is their model adequate to the desired answer? (3) Is the pattern only spatial (i.e. a common dipsersal route) or spatial and temporal (i.e. a common dispersal route, roughly at the same time)? (4) What about the taxa that departs from the selected model? Questions (1) is never answered by Sanmartín et al., question (2) is a clear no (the symmetrical model contradicts directly their aim), maybe they can give an answer to question (3) but they are more interested in parameter estimation than in biological questions, the question (4) is never answered, even when they accept that there are several different models proposed for different taxa in the Canarias.

[1] Goloboff, P.A., Pol, D. 2005. Parsimony and Bayesian phylogenetics. In: Albert, V.A. Ed. Parsimony, phylogeny, and genomics. Oxford univ., Oxford, pp. 148-159.