1st paper: evolution of vocal complexity

2 main goals:

1. Verbal-quantitative description of song structure for major clades (MARCELO DOES THIS)

  • Describe unique features of song structure for each major clade

  • Use random forest to determine song distinctiveness for each clade

2. Evaluate scenarios of song complexity evolution (always towards higher complexity?)

  • Determine the following song complexity parameters for each species, including swifts (outpgroup) (MARCELO DOES THIS):
    • Number of element types
    • Sequence complexity (index of number of transitions corrected by sequence length and number of element types)
    • Acoustic space (area occupied by the elements of a song in a random forest proximity space -after Multidimensional Scalling-)
  • Categorize repertoire size in 3 categories (low-medium-high) (MARCELO DOES THIS?)

  • Evaluate if transition where more likely towards increasing complexity (e.g. low-med and med-high higher than high-med and med-low) (DANIEL DOES THIS)

  • Use outgroup (swifts) only for reconstructing complexity, not for testing

Current state of the analysis

  • Song structure analyzed for 296 species (265 species & 31 swifts)
  • Complexity parameters were measured for all species
  • Around 15% of the hummingbird species were imputed from a birdtree phylogeny or assumed to be sister species for recent taxonomic splits
  • Posterior prob. hummingbird trees will be available soon if we want to account for phylogenetic uncertainty
  • Swift tree was obtained from birdtree (apparently no better phylogeny available) (consensus tree built using maximum credibility clade phangorn::maxCladeCred)

Current issues

  • Bind the two phylogenies (hummer and swifts) to a single ultrametric tree

  • Ideally a single complexity metric would be used for testing complexity evolution, however, as Daniel has pointed out, reconstructing PCA vectors could be misleading. Should we instead reconstruct each parameters independently and then calculate the complexity metric for each node and tip using reconstructed values??

  • How should we split the complexity metric (phyloPC1 on complexity parameters)? It seems that the way we split it can have an strong effect on the inference about complexity evolution. 2 objective ways could be: split the metric in 3 equal length intervals (0-0.33, 0.33-0.66, 0.66-1) or split it in 3 intervals with equal number of species (1/3 of species on each interval). Any thoughts on this are welcome!

  • Not sure how to account for phylogenetic uncertainty (or if we should at all)

  • We might have missed the ‘complex song’ of some species, so ideally we will run the analysis with and without those species

Additional thoughts

  • Can we use other phylogenetic techniques to come up with other parameters to describe the evolution of songs in the group and/or the particularities of each clade’s song structure?

NEXT STEPS (Daniel ideas)

  • Bind the two phylogenies (hummer and swifts) to a single ultrametric tree
  • Test for directional evolution using continuous complx metrics (and maybe a single combined metric)
  • Look at correlated evolution of different complexity metrics (e.g. does acoustic space and sequence complx evolve in a similar way)
  • play it be ear on whether to combine all 3 parameters of complexity and try evol. transition tests
  • build different hypothesis representing directional evolution in the Q matrix (e.g. all increasing rates equal and all decresing equal)
  • other option for transitions would be a bayesian approach to calculate breaks for categorizing complexity and calculate rates at the same time

Alternative/complementary tests

  • For discrete complexity use hidden states to see if transition rates change through time
  • Compare evolution of disparity among hummer clades