If you mean that strong assumptions of how evolution works are necessary for huge data sets, like those just published in Nature, I understand that point, but my point was about “most trees,” not even “most tree diagrams these days.” Moreover, it’s hard to intuit why any evolutionary assumptions would be necessary for constructing basic distance trees for orthologous or homologous protein sequences that demonstrate the common ancestry of humans and other great apes or primates.
And seeing nature as requiring God’s tinkering in the form of IDcreationism clearly has a high negative association with producing information with any practical value, in addition to diminishing the very idea of the Abrahamic God.
It’s another day without a creationist coronavirus vaccine…
BLAST is a way to search for sequences similar to a target. As such is produces a sort of genetic distance (though I’m not personally sure just what goes into that distance score). I didn’t know there was a widget to turn those distances into a tree. But what you have there is a minimum evolution method using BLAST distances. That’s a strong (and odd) assumption about how evolution works.
Do molecular biologists actually use BLAST to construct published trees? If so, I apologize for not knowing that. But I’m also appalled that anyone would do so rather than using a real phylogenetic analysis program. Even MEGA would be way better.
I, on the other hand, was talking about phylogenetic analyses. And those trees are usually produced these days by some kind of maximum likelhood algorithm, in which the model of evolution and its assumptions are very explicit.
No. The size of the data set isn’t relevant. They’re necessary if you want to produce a tree that reflects the data correctly. If you don’t care and just want something, anything, then sure, that BLAST widget would be fine.
Are you familar with, for example, Felsenstein 1978? There’s a huge literature on the inadequacy of simple methods, of which that’s probably the first. Those simple methods all made implicit assumptions about evolution, often very strong ones, that are seriously violated by real data. Yes, for some questions the data are good enough that any idiot analysis will produce a good tree. But it’s not the way to bet.
And I still maintain that most trees are not produced without explicit assumptions of how evolution works.
I think your discussion with @Mercer demands it’s own thread. I also thought BLAST was sufficient for constructing phylogenetic trees, but you seem to disagree. Maybe you could give a brief primer on the limitations of BLAST and other similar, “simple” programs for serious phylogenetic tree construction. The association of laymen would benefit from this.
First off, BLAST was never intended to do tree construction. Second, it uses a very simple algorithm, minimum evolution, that just doesn’t fit the way evolution works very well. It allows for another, neighbor joining, which is a quick and dirty approximation of a least squares fit. These both have two major problems: 1) they throw away the actual sequence and just use genetic distances; it’s not good to get rid of evidence. 2) they don’t test different trees against each other so can’t determine that one tree is better than another; they just go straight to a tree, and that’s it. Better algorithms 1) use the actual sequences and fit them to trees and 2) have optimality criteria by which one tree can be determined to fit the data better than another, and look at a great many trees to find the best one. PAUP is popular, and RaXML, and PHYLIP, and Mr.Bayes, and various others. All of them incorporate explicit evolutionary models. Nobody who does phylogenetic would ever use BLAST to build a tree, though apparently molecular biologists may. I wouldn’t credit such a tree.
BLAST is a way to search a sequence database for sequences that are similar to some reference sequence. Excellent way to search GenBank for homologous sequences. That’s what it was always intended for.
The topic has wandered, as topics generally do. The question is about how most trees are produced.
No indeed. That’s a common misconception. Colin Patterson, for example, was convinced that it was necessary in building trees to make no evolutionary assumptions, because only in that way could trees be used as evidence for evolution. But in fact his favored parsimony algorithm implicitly makes a number of assumptions about evolution; you can’t get away from them. Likelihood has the advantage that its assumptions are explicit and more in keeping with the nature of the data. Further, whether a tree is evidence for common descent depends on whether the tree is better supported by the data than other trees, and that’s easy enough to do in a likelihood framework.
Feel free to do poor analyses in preference to good ones, but it’s a bad habit to get into. And the resulting single, untested tree is not evidence for evolution.