Get best shift configuration does not work on combined eventdatas

[Jamie T]

Hi,

The tree I am using is so large it needed to be broken up and analysed with BAMM separately, then combined.  This works for rates through time and tip-rates, but when I try and run getbestshiftconfiguration it gives this error:

"Processing event data from data.frame

Discarded as burnin: GENERATIONS <  0
Analyzing  1  samples from posterior

Setting recursive sequence on tree...

Done with recursive sequence

Error in `[<-`(`*tmp*`, zzindex, 1, value = rep(k, length(zzindex))) :
  subscript out of bounds"

Another query I have is whether the expected number of rate shifts should be combined for this command.  I.e. if there are 4 individual analyses with an expected rate shift of 50 in each, should I specify 200 for analysis of the combined event data?

Cheers,

Jamie

[pasca...@gmail.com]

Hi Jamie,

Can you provide a bit more context? How was the tree broken up? Is there any overlap between subsets? How did you combine the output?

-Pascal

[Jamie T]

Hi Pascal,

It is the angiosperms tree analysed by Igea and Tanentzap (2020) (https://onlinelibrary.wiley.com/doi/full/10.1111/ele.13476).  They described what they did below, but I am not sure how they combined the event datas.  Certainly there is overlap.

Cheers,

Jamie

"Following Igea et al. 2017, convergence of such a large BAMM analysis was achieved by dividing the initial tree into clades of less than 6000 species. Doing so resulted in 17 clades (16 monophyletic clades and 1 clade that contained the backbone of the tree plus the unassigned species), which we then used to run independent BAMM speciation/extinction analyses. Within each clade, non-random incomplete taxon sampling was incorporated into the estimation process by calculating the number of species sampled in each family. Prior settings were obtained using the setBAMMpriors function in the R package BAMMtools (Rabosky et al. 2014) and expectedNumberOfShifts was set at 50. Analyses were run for a minimum of 100 million generations and ≥ 20% of each run was discarded as a burn-in when necessary. Effective samples sizes of the log-likelihood and the number of rate shifts were both above 200. The event files for all 17 clades were then combined to generate a single BAMM result file that was used for all downstream analyses."