Hacker News Reader

Comment by dekhn | Hacker News Reader

I worked in phylogenetics for a while and it was a pretty confusing area. Originally, phylogenetic trees (not the biological trees that are the subject of the OP) were created by finding physical features (yes, just like ML) and using those to build a semi-supervised tree-structure of classifications. However, eventually we began to use DNA sequences to compare organisms, which restructured the tree in many ways, even close to the root. It was a controversial time as the the historical physical-feature classifier group was certain their way was right, and same for the DNA folks. I sort of assumed that the DNA would be a much higher quality source for clustering but it hasn't really always worked out that way.

View on HN

I thought the article was going to be talking about phylogenetic trees aren't really trees (in the data structure sense) because the prevalence of horizontal gene transfer and hybridization helplessly muddies things.

I guess it depends on on why you are doing the clustering, right?

If you want to cluster animals that sort of fuzzily behave similarly together — if it looks like a duck and walks like a duck, it probably acts like a duck, haha.

The point of making trees is almost always to understand the fundamental sources of phenotypic diversity and making simple models that explain how the new (complex, organismal) phenotypes arise, how they are shared through various forms of reproduction and gene transfer, and how those relate to molecular phenotypes.

Nobody that I am aware of is clustering things based on similar functionality simply to say "look, these two things are similar", without having some sort of "explanation".

Except phylogeny generally was used to categorize how hereditarily related things were. Whether sequence alignment or physical taxonomy are better determinants is perhaps an open question in some cases, although I’m firmly on the side of sequence alignment, since DNA is an important component of inheritance

The side you are on has already won. Sequence data are universally accepted in systematics as the ultimate arbiters of evolutionary relationships, just delve into the modern classification of any extant group. One example of where molecular evidence overturned physical evidence long thought to be unambiguous, leading to a complete and essentially uncontested taxonomic overhaul: coprinoid mushrooms. Morphological characters are still used in paleontology for obvious reasons, but again within a higher-level framework provided by molecular phylogenetics.

Really? Cool! This was still pretty much unresolved in 2003 when I worked in the area and I had been told people were still arguing about this recently.

Are you proposing that we apply the notion of duck-typing to plants?

If it acts like a classification scheme, we should treat it as a classification scheme!

Please do elaborate for a non-biologist - how and when is DNA sequencing worse at phylogenetic grouping than physiology? I had assumed the latter was embarrassingly outdated and easily duped (by convergent evolution, etc)

It's been about 20 years since I last looked into this but I think there were a few cases where people had "golden classifications" (IE, some sort of external proof of the grouping) that were more consistent with character features.

To be honest I'm not the best person to ask because every time I dip my toes in the area I realize (a) how little I know and (b) just how ugly these debates get. and (c) how much scientists like to treat some side observation as golden data that is absolutely right when trying to build support for their theory

Reply on news.ycombinator.com