Hi Belinda, Sorry to jump in on this a bit late, but I thought maybe we need to step back to basics first?
Since lambda=~0, this indicates strongly that your data has almost no phylogenetic signal; most of the variation in data is at the tips = noisy with respect to phylogeny. In which case, I find it difficult to imagine a homogenous OU process being able to explain such highly variable data at the tips. If you fit two badly fitting models, you'd likely end up with one fairing better than the other, in your case, OU >> BM. But you should check that your OU model is a good fit to your data, which I suspect is not. This may be a silly question but have you logged your data? It is common to see very low lambda when data is not logged. thanks, Manabu On 15 April 2016 at 20:59, Hunt, Gene <hu...@si.edu> wrote: > Hi Belinda, > > I agree with Brian’s comment. I’d also add the more general point that OU > models don’t really have a clearly interpretable rate parameter in the same > way that BM models do. In BM models, the sigma parameter controls the pace > of trait evolution: the expected change over a given time interval depends > strightforwardly on sigma (and only on sigma). > > OU models also have a sigma parameter that has the same meaning — it is > the variance of the diffusion process. But, it is perilous to interpret it > in the OU context as a rate parameter as one would for BM. This is because > the expected change over a given interval depends on sigma, but also on > alpha and the distance to the optimum. This complexity means that OU > models don’t have any single number you can point to as a generic rate of > change for the process. > > So, if your scientific question hinges on saying something about rates on > branches, I think this is all the more reason to follow Brian’s suggestion > of looking into one of the BM rate heterogeneity methods. > > Best, > Gene > > -- > Gene Hunt > Curator, Department of Paleobiology > National Museum of Natural History > Smithsonian Institution [NHB, MRC 121] > P.O. Box 37012 > Washington DC 20013-7012 > Phone: 202-633-1331 Fax: 202-786-2832 > http://paleobiology.si.edu/staff/individuals/hunt.cfm > > From: Brian O'Meara <omeara.br...@gmail.com<mailto:omeara.br...@gmail.com > >> > Reply-To: "omeara.br...@gmail.com<mailto:omeara.br...@gmail.com>" < > omeara.br...@gmail.com<mailto:omeara.br...@gmail.com>> > Date: Friday, April 15, 2016 at 11:00 AM > To: Belinda Kahnt <belind...@gmx.de<mailto:belind...@gmx.de>>, "mailman, > r-sig-phylo" <r-sig-phylo@r-project.org<mailto:r-sig-phylo@r-project.org>> > Subject: Re: [R-sig-phylo] estimating the evolutionary rate of a continous > trait > > Hi, Belinda. > > One thing to watch is over-intepretation: BM is consistent with drift, OU > is consistent with selection, but various kinds of selection can also lead > to BM. [1]. A lot of the people involved in these methods (including me) > are guilty of sloppiness in language in this area, leading to confusion. > > Also, another issue could be branch lengths: are they proportional to time > (or, arguably, something like number of generations)? All these methods are > basically stretching the tree in various ways (and, for OU, adjusting > expected means), so if branch lengths are arbitrary, so are the results. I > ask due to the lambda fit. Other things that can cause this are > unincorporated measurement error (b/c it looks like a lot of evolution > right at the tips, meaning little evolution along the branches). I'd > suggest incorporating this (you can give known estimates of measurement > error; some programs allow this to be estimated as well (I forget whether > we have the estimation bit exposed in OUwie, but known error should be > there at least)). > > Surface can estimate different regimes on the tree, but IIRC it does not > estimate different rates, just OU means. I believe auteur (in geiger) and > BAMM can model different BM rates on different branches, which sounds like > your question. We have some code to try different OU models (including > perhaps ones with different rates) on different branches buried somewhere > in OUwie (we call it OUwie-dredge so people know to be cautious) but it > hasn't been tested or published yet. > > One other caution: if you have a BM model, and are just modeling different > rates, it can be done well. Trying to estimate different rates with OU > models is harder to do well: there's interaction between alpha and sigma > that can make them difficult to distinguish (not formally nonidentifiable > [unlike OU means sometimes], just difficult). For me, I'd probably lean > towards one of the BM rate heterogeneity methods for your question. > > Hope this helps, > Brian > > > [1] Hansen, T. F. and E. P. Martins. 1996. Translating between > microevolutionary process and macroevolutionary patterns: the correlation > structure of interspecific data. Evolution 50:1404-1417. > > _______________________________________ > Brian O'Meara > Associate Professor > Dept. of Ecology & Evolutionary Biology > U. of Tennessee, Knoxville > http://www.brianomeara.info > > > > On Fri, Apr 15, 2016 at 10:20 AM, Belinda Kahnt <belind...@gmx.de<mailto: > belind...@gmx.de>> wrote: > > Dear all, > I am a newbie to this mailing list and phylogenetic analyses in R and hope > you can comment on a question I have. I would like to estimate the rate of > evolution of a continous trait and check if the trait evolves faster along > some branches of the topology. I already checked with Pagels lambda if the > trait evolves according to a Brownian motion model,(i.e. pure drift), which > it did not (lambda ~ 0). Modelling the trait evolution under the > Ornstein-Uhlenbeck (OU) model provided a significant better fit (p<- 0.01) > with a very high alpha parameter of 8.7 (indicating a role of selection in > trait evolution). In order to infer now the rate of trait evolution I am > searching for a R package that is able to do this inference based on the OU > model. However, so far I either just found methods based on the Brownian > motion model (e.g. brownie.lite function in phytools, Motmot) or methods > that also rely on the reconstruction of ancestral selective regimes > beforehand (like mvMORPH or OUwie). I want to keep the model as simple as > possible and therefore don't know if it is necessary to also model > ancestral selective regimes if I just want to know if the evolutionary rate > of the trait varies along branches and which branches show an accelerated > rate?! Which programm would you recommend? I would be very thankful for > your help and recommendations. > Best wishes, > Belinda > > _______________________________________________ > R-sig-phylo mailing list - R-sig-phylo@r-project.org<mailto: > R-sig-phylo@r-project.org> > https://stat.ethz.ch/mailman/listinfo/r-sig-phylo > Searchable archive at > http://www.mail-archive.com/r-sig-phylo@r-project.org/ > > > [[alternative HTML version deleted]] > > _______________________________________________ > R-sig-phylo mailing list - R-sig-phylo@r-project.org<mailto: > R-sig-phylo@r-project.org> > https://stat.ethz.ch/mailman/listinfo/r-sig-phylo > Searchable archive at > http://www.mail-archive.com/r-sig-phylo@r-project.org/ > > > [[alternative HTML version deleted]] > > > _______________________________________________ > R-sig-phylo mailing list - R-sig-phylo@r-project.org > https://stat.ethz.ch/mailman/listinfo/r-sig-phylo > Searchable archive at > http://www.mail-archive.com/r-sig-phylo@r-project.org/ > -- Manabu Sakamoto, PhD manabu.sakam...@gmail.com [[alternative HTML version deleted]] _______________________________________________ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
