Friday, February 5, 2016

#ParsimonyGate: The Perspective of a Reformed ‘Hardcore’ Cladist


If you are reading this article you have probably read the now infamous editorial in the journal Cladistics http://onlinelibrary.wiley.com/doi/10.1111/cla.12148/full . Although signed “The Editors,” it isn’t clear if this was approved by anyone besides the current Head Editor who is most certainly a “hardcore cladist” (someone who thinks parsimony is the most reasonable, if not only, tool for inferring the historical relationships of organisms through phylogenetics). I distinguish between “hardcore cladists” and just “cladists,” because I think I am a cladist and that most other systematic biologists are too. A cladist in my personal definition is anyone who is distinguishing between pleisiomorphic (“primitive” features shared with a designated outgroup) versus apomorphic (derived characters distinct from the outgroup condition). Under my broader definition, basically everyone doing morphological or molecular work to discover the relationships of organisms is a cladist, and it doesn't matter if you are using parsimony, likelihood, or Bayesian approaches. The only exception are folks that are using overall similarity (e.g., bats and birds are close relatives because they both have warm blood and wings) which doesn’t distinguish analogy (convergence of characters) versus homology (characters derived from common ancestry) because it doesn’t follow the Henningian, or cladistic principal, of distinguishing between pleisiomorphic versus apomorphic characters.
            This view of cladistics I outline above are basically the foundation of Willi Hennig’s 1966 book “Phylogenetic Systematics” that is the bible of the Willi Hennig Society (publisher of Cladistics) and the foundation of modern systematic theory. At the time the idea of distinguishing between pleisiomorphy versus apomorphy was radical. Famed evolutionary biologist Ernst Mayr was the first to call those following Hennig’s principles “cladists” - as a pejorative by the way. Mayr preferred doing “evolutionary taxonomy” - basically where the expert on a group makes a hypothesis about the relationships of organisms based on characters they think are most important for supporting those relationships (e.g., owls, eagles and hawks are all each other’s closest relatives because these “raptors” all kill with their feet). The other alternative method in systematics in those early days were the numerical pheneticists that used overall similarity to group organisms as I explain above. (Read more about this interesting time in history in David Hull’s, “Science as a Process.”) The original cladists weren’t fighting for parsimony, they were fighting to only use derived characters in phylogenetics. Parsimony came around a little later with the work of several groups mainly from the University of Michigan and American Museum of Natural History. Parsimony was the only game in town to the early cladists, which was mainly for understanding the transition of morphological characters from primitive to derived. Then with the rise of molecular tools for obtaining DNA characters came new methods for inferring trees: model-based approaches including maximum likelihood and eventually Bayesian inference. Systematists of all sorts would meet at the annual Systematic Zoology/Biology meetings every year until the “hardcore cladists” decided to break away and have their own meeting, the meeting of the Willi Hennig Society founded in 1980.
            Now I should mention I trained as a systematist at both the University of Michigan and the American Museum of Natural History, the hot bed of cladistics and Hennig worship, albeit late in the game in the early 2000s. I was a hardcore cladist most of my early graduate career. I thought parsimony was the only reasonable way to infer relationships because it wasn’t a model-based approach like maximum likelihood or Bayesian inference. Those models made too many assumptions I thought and was taught. Alternatively, parsimony wasn’t a model because the foundation of that idea is to “minimize ad hoc assumptions about homoplasy” (i.e., reduce noise in the tree from characters moving around). Using parsimony, the shortest tree with the fewest steps (or evolutionary transitions) is the best tree – period. The other methods were using models to guesstimate from DNA sequences too much about how often an A (adenine) turns to a C (cytosine) or a T (thymine) to a G (guanine). It was crazy how much assuming those crazy-assuming people were doing. If they just did some morphology they would better understand how all this stuff really worked and that there is only one true religion, I mean method, parsimony. We were the Jedi knights that stuck to our principles; those other folks just weren’t thinking it through. Then something happened: I saw the light.
            I realized at one point that there isn’t a right way to study historical relationships. We can’t actually know the truth about who is related to whom when discussing organisms that diverged millions of years ago. We are also using methods that are extremely computationally intensive. They are all models, even the heuristic we use to run parsimony. No computer on Earth can fully resolve a phylogeny of more than a dozen or so species using any heuristic of parsimony or likelihood: there are just too many possible answers. When we study a historical science using morphological characters or DNA we will never be sure we are right. As I started using DNA methods more I realized I wanted to start better understanding when these lineages started to diverge. I needed to put a rough age on a group and to do that I needed to use likelihood and Bayes because only those use evolutionary models from which you can understand how DNA sequences change over time. I slowly found myself using these other methods more and more. Did I still use parsimony, sure sometimes, but it gave me the same answer as those other methods, just less information (e.g., a tree without branch lengths or information about time). The relationships themselves are interesting but I also wanted to know about evolution and biogeography beyond the tree.
            Now I’m still a cladist, and I hope I can count many friends among those in the Willi Hennig Society. (I named my dog Willi.) The folks in that society helped me think more clearly about methods and the philosophy of systematics, and also about the limits of what we can know in general (epistemology). They have invited me for talks at their annual Hennig meetings and I always learn a lot at this conference. Many people are intimidated by these meetings because many senior members do yell at each other, but they are friends in the end - trying to improve each others work. They do sometimes pick on folks that aren’t their friends, and that isn’t cool. You do have to bring your “A” game to Hennig because there are no concurrent sessions and there is an unlimited time for questions. You always have to explain why you picked a certain method over another, it isn’t about using the newest method it is about justifying your choice. (Much like the Cladistics editorial was trying to say. I think.) Compared to other meetings where there is often few, if any, questions - even after a terrible talk - I actually think Hennig is doing it right. They have many fewer members than other major systematic societies so they have the luxury of having just one session at a time and an open-ended question period. The Hennig conference is also strongly skewed male, which is a problem they really need to fix. Many senior members of the society need to tone it down a bit too. They can be crass and pedantic and use jargon as a weapon to make semantic arguments over relatively mundane things (“how can you test a model with a model”; “is there such a thing as an order-quantifiable metric of similarity”). I still publish in Cladistics (as recently as last year) and it even had a Bayesian analysis in it. Although I let my membership lapse a few years ago I’m not opposed to going to another meeting in the future. I think the editorial they published is a step backwards only because it sounds so uninviting: “If alternative methods give different results and the author prefers an unparsimonious topology, he or she is welcome to present that result, but should be prepared to defend it on philosophical grounds.” Many read that as, “You can submit non-parsimony things but you need to explain why, and even if you explain why, we still might not like it because parsimony.”
            I think the editorial was a mistake because it sounded like they will only accept the parsimony answer if you get alternatives from other sources. And that makes the journal “Hardcore Cladistics” and it was, at least recently, just “Cladistics.” I do hope they reconsider their stance, or at least clarify. I still consider Cladistics a great journal, one that I enjoy reading because of its organismal focus on systematics. I haven’t had issues with editors or reviewers telling me I need to do a parsimony analysis or remove a likelihood or Bayesian analysis, but I’ve heard that other may have. Time will tell if the journal and the society can right the ship, unfortunately, it was a storm of their own creation that has it teetering.