I let the experts answering Jacqueline's questions, but have a related one. It's not about phylogenetic inference using morphometrics.
I've reviewed a few paleoanthropological papers where the authors used a phylogeny that included fossil hominins as the backbone to apply comparative methods to cranial shape data. At least for the fossils (with few exceptions like Neanderthals, where ancient DNA is available), the phylogeny was a morphological analysis using mostly cranial data. The phylogeny was usually taken from the literature and, therefore, was not using the same shape measurements as in the study but was still based on cranial data (meristic, traditional morphometric etc.). My guess is that this is a mistake because the data (cranial morphology, even if not exactly the same variables) are not independent. This is not specific to paleoanthropological data and probably a common issue in the analysis of fossils with no independent (like a molecular phylogeny for living species) information. Am I off track? Cheers Andrea ---------- Forwarded message --------- From: Jacqueline Silviria <[email protected]> Date: Thu, 8 Feb 2024 at 01:46 Subject: Re: [MORPHMET2] Compiling metadata on all methods of phylogenetic inference using GM data To: Joe Felsenstein <[email protected]> Cc: Chris Klingenberg <[email protected]>, Joe Felsenstein < [email protected]>, andrea cardini <[email protected]>, [email protected] <[email protected]> Good evening, Apologies for the delayed response. Contml was originally designed for inferring phylogenies from gene frequency data in closely related populations, under the assumption that the mechanism of change is genetic drift. Then it is roughly possible to assume independent Brownian Motions of allele frequencies, with equal variances of change. For quantitative characters all these assumptions are highly dubious. I try to point out to the questioner that they cannot assume that the characters are independent, and that they change with equal variances. The questioners rarely react with enthusiasm to this. This appears to be *the* issue with all proposed methods for analyzing GM data, save perhaps the PC-based methods. Proposed parsimony-based methods of accounting for inapplicability and interdependence in discrete characters (e.g. Brazeau *et al.*, 2019 <https://academic.oup.com/sysbio/article/68/4/619/5238046>; Hopkins & St. John, 2021 <https://academic.oup.com/sysbio/article/70/6/1163/6131693>; Goloboff* et al.*, 2021 <https://onlinelibrary.wiley.com/doi/full/10.1111/cla.12456>; Wheeler, 2023 <https://onlinelibrary.wiley.com/doi/full/10.1111/cla.12553>; Goloboff & De Laet, in press <https://onlinelibrary.wiley.com/doi/full/10.1111/cla.12564>) aren’t easily applicable to continuously varying data (see review by Goloboff, 2022), and I’ve yet to encounter a likelihoodist alternative. I would encourage people to try going the PCM route: getting a tree for living species and using it to infer covariances of changes in characters. I have mentioned in a few places that from that one might even use the resulting inference to place fossil species on a tree (perhaps Liam Revell has published that suggestion too). I’m assuming you mean Revell *et al.* (2015) <https://academic.oup.com/evolut/article/69/4/1027/6852435>. I imagine this procedure is easily workable when the molecular “backbone” topology is robustly supported, and the developmental constraints of morphometric characters are well-known for both extant and extinct taxa, e.g., size and shape of mammalian teeth. I don’t know if it’s workable for taxa without any molecular dats, without close modern relatives, and with non-modern-analog morphology and ontogeny, e.g., graptolites or vendobionts. I would also note that there is a program, Threshml, distributed by my lab, that uses Sewall Wright's "threshold model" to model discrete character evolution, and to treat both continuous and discrete characters in the same analysis. I did not know this. Thanks! *Jacqueline S. SilviriaThe Last King of the JungleDepartment of Earth & Space ScienceUniversity of WashingtonSeattle, WA, [email protected] <[email protected]>, [email protected] <[email protected]>* *ResearchGate profile <https://www.researchgate.net/profile/J_Silviria>* *Twitter: @JSilviria* Sent from my iPhone On Feb 5, 2024, at 6:33 PM, Joe Felsenstein <[email protected]> wrote: Morphmet people -- I want to concur with Jacqueline Silviria and with Chris Klingenberg about the difficulties of using morphometric analyses to infer phylogenies. Just to "violently agree": This is a more general problem with quantitative characters of any kind. I often get asked how to use the continuous-character maximum likelihood program Contml in my package PHYLIP to infer a phylogeny from quantitative characters such as morphological measurements. Contml was originally designed for inferring phylogenies from gene frequency data in closely related populations, under the assumption that the mechanism of change is genetic drift. Then it is roughly possible to assume independent Brownian Motions of allele frequencies, with equal variances of change. For quantitative characters all these assumptions are highly dubious. I try to point out to the questioner that they cannot assume that the characters are independent, and that they change with equal variances. The questioners rarely react with enthusiasm to this. Of course there is a literature (Phylogenetic Comparative Methods) where we have methods (which I guess we should call "Phylogenetic Comparative Methods methods") that take a tree that is presumed known, or a cloud of trees from bootstrapping or a sampled Bayesian posterior, and work the other way. Namely they infer the covariances between changes in a multivariate Brownian Motion, given the tree. I know that in the brilliant theory of Kendall, there are theorems showing that if the individual landmark (x,y) coordinates have equal and isotropic i.i.d. variances, then so will the coordinates in the shape space. But what that is doing is assuming that the differences we see are all measurement error. It does not guarantee independent i.i.d. Brownian Motion when the changes along the branches of the tree are themselves covarying among characters and with unequal variances. I would encourage people to try going the PCM route: getting a tree for living species and using it to infer covariances of changes in characters. I have mentioned in a few places that from that one might even use the resulting inference to place fossil species on a tree (perhaps Liam Revell has published that suggestion too). I would also note that there is a program, Threshml, distributed by my lab, that uses Sewall Wright's "threshold model" to model discrete character evolution, and to treat both continuous and discrete characters in the same analysis. Joe ---- Joe Felsenstein [email protected], [email protected] <[email protected]> Department of Genome Sciences and Department of Biology, University of Washington, Box 355065, Seattle, WA 98195-5065 USA ----- PS: please do not use [email protected], which is an alias that some mail systems now mistake as indicating spam. -- E-mail address: [email protected], [email protected] WEBPAGE: https://sites.google.com/view/alcardini2/ or https://tinyurl.com/andreacardini -- You received this message because you are subscribed to the Google Groups "Morphmet" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/morphmet2/CAJ__j7MuNodkm_hg4ewiRvJffZWBKSkSB2hN9e4KbHNEGovCMw%40mail.gmail.com.
