Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Researchers use genomes of 241 species to redefine mammalian tree of life (phys.org)
106 points by wglb on May 14, 2023 | hide | past | favorite | 20 comments


https://www.science.org/doi/10.1126/science.abl8189

It looks like they looked at a larger amount of the genome, and still found that relationships at the level of Order and below were hard to be sure about, since different genomic regions disagreed. They then speculate that this is related to different geological / tectonic / geomorphological processes occurring at that point in history, compared to the time of the earlier divergences. It’s a frustrating area of science; we’ve learned that the reality is that some aspects of evolutionary history are very hard to be sure about, but scientific publishing and careers require more interesting claims.


Euarchontoglira would be a way less researched and relevant branch of life have it not produced us humans.

Same thing that if sentient birds were to appear, they will consider dinosaur epoch never ended, just had a setback and useful bottleneck.


>> sentient birds

have you seen the problem solving of crows?

edit: vid: https://www.youtube.com/watch?v=Gui3IswQ0DI


Where can I read any science written by crows?


I think you’re not using “sentience” in the same way as each other.

I recommend avoiding “sentience” in this thread and being more specific about what you mean.


>> way less researched and relevant branch of life have it not produced us humans.

I doubt that any branch would have been even studied in any way had not euarchontoglira produced us humans. I doubt many other animals are even aware of evolution as a concept.


Are African Greys sentient bird examples?


Did they manage to build and publish an evolution tree?


I think that’s beyond the meaning of sentient.


"sentient" is often used casually when "sapient" is meant

don't whether that's country-level language variation, an example of language evolution ;), or just sloppiness.


"Sapient" is a stupid term because a Denisovian or Neanderthal were likely able to produce a piece of research with some assistance, while being a Homo {not Sapient}.


So horses are closer to lions than to antelopes? The pictures on phys.org and in the Science article preview are frustratingly unclear; most leaves of the tree are unlabelled.


Sort of - it's more accurate to say that the order Perissodactyla (containing horses) shares a common ancestor with the order carnivora labeled #4 in the picture. Both of them share a common ancestor with artiodactyla (even toed ungulates like antelope) labeled #3.

They all differentiated over 60 million years ago, though, which was enough time for mammals to split into herbivores and carnivores and evolve back into sea animals like whales and dolphins. With converging and diverging evolution over that kind of time span, saying an animal is "closer to" another over simplifies a lot of detail. One species' genetics might resemble one order while the slowly evolving genes used to identify evolutionary relationships place it in a second order.


I think the layout is optimized for showing branches event and not relative difference between species, though as a side effect of showing branching it does somewhat approximate relative difference.


A flip side of constraint is constrained diversity. Does a tree always work? Where are the genomes in ILS? Those will be very interesting regions.


Trees are tricky. They assume solid splits between species. While that may work well across millions/billions of years it does not work well when describing the specifics of split points. There is no bright line when one generation is suddenly a different species than the last. Genetics describes the slow accumulation of changes. Translating that into a tree of lines connected by points will always require compromise.


From the abstract: "However, because these approaches infer a single best tree, they mask signatures of phylogenetic conflict that result from incomplete lineage sorting and historical hybridization."

So it can't be avoided, and the solution of making more trees from different stuff is somewhat unsatisfying. They have to work with the methods they have today.


They used maximum likelihood tree building. In Bayesian approach it’s possible to sample from a distribution of trees which can help with the uncertainty. Anyway, I only have very limited knowledge on this from my evolutionary biology course.


Reading the paper I think they used both SVDquartets (coalescence-based explicitly for ILS) and concatenation + maximum likelihood, which is a very sane combination for the analysis. Not contradicting you here, but at their scale it is unlikely any Bayesian approach will converge, so they had to settle with whatever they had to quantify uncertainty (branch supports and others).

Edit: and also the Bayesian approaches might have model mismatch, for example I am not sure what's an accurate Bayesian method that models hybridization and ILS


Building evolutionary trees from genomes is challenging, in part because of the difficulties inferring trees when different parts of the genome may be evolving at very different rates, but also because of the wide dynamic range of genome quality. Human and mouse genomes have far fewer errors (missing regions) than rat and cow, and many of the other several hundred genomes will be of much much lower quality. At this scale, it is difficult to know whether surprising results reflect biology or inaccuracy.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: