Pipeline publication

[…] nd our results remain conditional on future denser sampling of the angiosperm phylogeny. Our strict exemplar approach also means that data are missing for some traits in some species (total missing data: 27%, including cases of inapplicability). Because missing or inapplicable data are more or less evenly and haphazardly distributed across our tree, and species with such data are in effect pruned out in the ancestral reconstruction analyses, it is unlikely that missing data had a strong impact on our results., Each floral trait was analysed for each series of trees (A, B, C, D, E, A200, B200, C200, D200, E200) using three complementary approaches: MP using the ancestral.pars function of the phangorn 2.0.2 package in R, ML using the rayDISC function of the corHMM 1.18 package in R, and a Bayesian rjMCMC approach using BayesTraits 2 (ref. ). MP and ML reconstructions were conducted on the MCC tree from each BEAST analysis, whereas Bayesian rjMCMC analyses were conducted on collections of at least 1,000 trees sampled from the posterior stationary distribution from the BEAST analyses. Here, we focus on and report results for 15 key nodes in the phylogeny of angiosperms, corresponding to well-recognized major clades (including Angiospermae, Mesangiospermae, Magnoliidae, Monocotyledoneae, Eudicotyledoneae, Pentapetalae, Rosidae and Asteridae). However, graphical MP and ML reconstructions for the entire tree are available ()., For each floral trait, we tested and compared at least two distinct Markov models of discrete character evolution in our ML analyses: the equal rates (ER) or Mk model, which assumes a single rate of transition among all possible states, and the all rates different (ARD) or AsymmMk model, which allows a distinct rate for each possible tr […]

Pipeline specifications

Software tools Phangorn, BayesTraits, BEAST