3 ms·

There is always going to be difficulty in trying to put categorical labels on continuous distributions. I've long wondered whether abandoning the concept of "s

There is always going to be difficulty in trying to put categorical labels on continuous distributions. I've long wondered whether abandoning the concept of "species" in favor of some continuous measure of genetic relatedness would work. From a practical, everyday standpoint, it would obviously be unworkable. We created the idea of species because in day-to-day life it's incredibly useful and only breaks down in some pretty niche edge cases (again: from the perspective of a lay person, those edge cases are both more common and more important to the scientists working with them). But as far as scientists, and particularly phylogenists and evolutionary biologists/ecologists go, it seems like we might save, if nothing else, a ton of effort that currently goes into trying to fit the round peg of nature into the square hole of our categorical system. That being said, phylogeny is at best only tangentially related to my own expertise so I'm sure there would be difficulties that I haven't though of.

kjkjadksj 3y ago
Most people use genetic relatedness when constructing phylogenies today. People in the field understand species labels are just discrete terms. Some fields like metagenomics don’t even consider the species much and group things by genus or some higher level instead because of the difficulty of defining this stuff. Not to mentioned most of what you sample using metagenomics is going to be a population pool (think someone extracting dna out of soil or stool, there’s billions of organisms you are sampling potentially in that one sample). You start to think less of “do i have this organism and what does this mean” and more “what gene sequences am I detecting out of this pool and what does this mean for the resulting proteins being released in the environment and what that does”
- ak217 3y ago
  Yep - metagenomics/phylogenomics practitioners have learned the hard way that trying to focus on species definitions gets you nowhere (if you think defining a species in multicellular sexually reproducing eukaryotes is hard, wait till you see what bacteria do - it's basically a free-for-all; and on top of that, short read metagenetic DNA sequencing gives you a soup of DNA that is very hard to precisely partition by source). I consider the focus on species and speciation to be a science social phenomenon more than anything else, a lot of it attributable to the prestige of being the one to "discover" the species (a.k.a. describe it in a way that sticks). The flipside is that the science of phylogenomics is incredibly fruitful, in that as we collect more complete information about genomes and their diversity and patterns of conservation, we are able to better understand life in all of its forms. It's also very data intensive and computationally complex - some of the most sophisticated data compression, suffix tree construction, hashing, and clustering algorithms are used in this field. An unsung group of heroes in the face of this insane complexity is the scientists at NCBI whose job it is to organize all of the taxonomic and reference genetic/genomic sequence information (they are the ones who build the GenBank, RefSeq, and NCBI Taxonomy databases). Their work is as close as it comes to the source of truth on this. While the GenBank FTP server is easily overwhelmed by modern workloads trying to ingest this data, AWS and GCP provide high throughput mirrors (https://aws.amazon.com/opendata/ https://aws.amazon.com/opendata/, s3://ncbi-blast-databases, gs://blast-db).
vrosas 3y ago
Humans have an inherent need to label, categorized and segregate things into discrete buckets that I find fascinating.
- zmgsabst 3y ago
  Because it’s highly effective. You just slap a topology on things and you get an effective type theory that gets mostly good results which allow you to elide most of the complexity. This in turn allows you to reason about more complex things at a loss of fidelity — what we call abstraction. If you think being able to recognize a forest rather than be overwhelmed by the number of leaves on trees is useful, then you understand why humans do that.
  - vrosas 3y ago
    If you’re not a bot say potato.
- lo_zamoyski 3y ago
  Says the person who has uttered a sentence chock full of categories! The "what" of something is part and parcel of intelligibility. If reality is intelligible, and we are truly rational creatures that are able to know this reality, then this would entail that reality is divided into kinds that we can recognize. What those kinds are, and which things belong to which kinds, is a separate question, indeed the question in this article as it relates to living things.
bongodongobob 3y ago
I wonder if the Hamming distance between the DNA sequences would be useful or even good enough. Maybe there's a threshold where over a certain distance, two organisms wouldn't be able to reproduce.
- melagonster 3y ago
  yes, this calls DNA barcode gap. but a lot of species don't care this and it just a empirical data.
- Someone 3y ago
  DNA base pair counts vary between species, so some edit distance (https://en.wikipedia.org/wiki/Edit_distance https://en.wikipedia.org/wiki/Edit_distance) would be better. You’d have to solve the problem that chromosome counts vary, too, though (in animals from 1 for the male jack jumper ant to over 500 in the Atlas blue butterfly. See https://en.wikipedia.org/wiki/List_of_organisms_by_chromosome_count https://en.wikipedia.org/wiki/List_of_organisms_by_chromosom...)
runeblaze 3y ago
I work on computational phylogenetics. A practical problem is that while we have decent models of many things, gene transfer, hybridization, etc., practical methods tend to focus on the simpler models. I recall that a popular reason why phylogenetic trees are AFAIK more popular than phylogenetic networks (despite networks being more powerful) is because simpler models lead to more effective algorithms that do well under the oversimplified-but-probably-good-enough models. Complicated models are much more intractable (well, simple models usually lead to NP-hard problems. Complicated models are... even more hairy). Also theory is much easier understood on simpler models. Just like big O notation abstracts over many things, simple models are much easier to reason about. There is a time and place for considering the memory hierarchy, how floating point division is really not the same level of speed as integer addition, but inferring the canine phylogeny under some easy-to-reason-about model of evolution is much simpler (both theoretically and practically), especially if you have already gathered data for five or six years and people are getting tired of no results.

kjkjadksj 3y ago

Most people use genetic relatedness when constructing phylogenies today. People in the field understand species labels are just discrete terms. Some fields like metagenomics don’t even consider the species much and group things by genus or some higher level instead because of the difficulty of defining this stuff. Not to mentioned most of what you sample using metagenomics is going to be a population pool (think someone extracting dna out of soil or stool, there’s billions of organisms you are sampling potentially in that one sample). You start to think less of “do i have this organism and what does this mean” and more “what gene sequences am I detecting out of this pool and what does this mean for the resulting proteins being released in the environment and what that does”

ak217 3y ago
Yep - metagenomics/phylogenomics practitioners have learned the hard way that trying to focus on species definitions gets you nowhere (if you think defining a species in multicellular sexually reproducing eukaryotes is hard, wait till you see what bacteria do - it's basically a free-for-all; and on top of that, short read metagenetic DNA sequencing gives you a soup of DNA that is very hard to precisely partition by source). I consider the focus on species and speciation to be a science social phenomenon more than anything else, a lot of it attributable to the prestige of being the one to "discover" the species (a.k.a. describe it in a way that sticks). The flipside is that the science of phylogenomics is incredibly fruitful, in that as we collect more complete information about genomes and their diversity and patterns of conservation, we are able to better understand life in all of its forms. It's also very data intensive and computationally complex - some of the most sophisticated data compression, suffix tree construction, hashing, and clustering algorithms are used in this field. An unsung group of heroes in the face of this insane complexity is the scientists at NCBI whose job it is to organize all of the taxonomic and reference genetic/genomic sequence information (they are the ones who build the GenBank, RefSeq, and NCBI Taxonomy databases). Their work is as close as it comes to the source of truth on this. While the GenBank FTP server is easily overwhelmed by modern workloads trying to ingest this data, AWS and GCP provide high throughput mirrors (https://aws.amazon.com/opendata/ https://aws.amazon.com/opendata/, s3://ncbi-blast-databases, gs://blast-db).

vrosas 3y ago

Humans have an inherent need to label, categorized and segregate things into discrete buckets that I find fascinating.

zmgsabst 3y ago
Because it’s highly effective. You just slap a topology on things and you get an effective type theory that gets mostly good results which allow you to elide most of the complexity. This in turn allows you to reason about more complex things at a loss of fidelity — what we call abstraction. If you think being able to recognize a forest rather than be overwhelmed by the number of leaves on trees is useful, then you understand why humans do that.
- vrosas 3y ago
  If you’re not a bot say potato.
lo_zamoyski 3y ago
Says the person who has uttered a sentence chock full of categories! The "what" of something is part and parcel of intelligibility. If reality is intelligible, and we are truly rational creatures that are able to know this reality, then this would entail that reality is divided into kinds that we can recognize. What those kinds are, and which things belong to which kinds, is a separate question, indeed the question in this article as it relates to living things.

bongodongobob 3y ago

I wonder if the Hamming distance between the DNA sequences would be useful or even good enough. Maybe there's a threshold where over a certain distance, two organisms wouldn't be able to reproduce.

melagonster 3y ago
yes, this calls DNA barcode gap. but a lot of species don't care this and it just a empirical data.
Someone 3y ago
DNA base pair counts vary between species, so some edit distance (https://en.wikipedia.org/wiki/Edit_distance https://en.wikipedia.org/wiki/Edit_distance) would be better. You’d have to solve the problem that chromosome counts vary, too, though (in animals from 1 for the male jack jumper ant to over 500 in the Atlas blue butterfly. See https://en.wikipedia.org/wiki/List_of_organisms_by_chromosome_count https://en.wikipedia.org/wiki/List_of_organisms_by_chromosom...)

runeblaze 3y ago

I work on computational phylogenetics. A practical problem is that while we have decent models of many things, gene transfer, hybridization, etc., practical methods tend to focus on the simpler models. I recall that a popular reason why phylogenetic trees are AFAIK more popular than phylogenetic networks (despite networks being more powerful) is because simpler models lead to more effective algorithms that do well under the oversimplified-but-probably-good-enough models. Complicated models are much more intractable (well, simple models usually lead to NP-hard problems. Complicated models are... even more hairy). Also theory is much easier understood on simpler models. Just like big O notation abstracts over many things, simple models are much easier to reason about. There is a time and place for considering the memory hierarchy, how floating point division is really not the same level of speed as integer addition, but inferring the canine phylogeny under some easy-to-reason-about model of evolution is much simpler (both theoretically and practically), especially if you have already gathered data for five or six years and people are getting tired of no results.