2.4.2 Bayes Factor species delimitation implemented in SNAPP
As an alternative to BPP+gdi , we used Bayes factor species
delimitation (BFD*) analyses in SNAPP v.1.3 (Bryant et al. 2012). For
the SNAPP analyses, 500 variable sites evenly distributed across the
dataset were extracted using a custom script and uploaded to a SNAPP
template in BEAUTi v.2.6.3 (Bouckaert et al., 2019). Alternative species
models were subsequently assigned based on prior species delimitation
inferences from ASAP, BPP, and well-supported clades inferred from in
the ML analyses (see results) recovered within distinct ASAP partitions.
Due to computational limitations in SNAPP, we divided the ML tree into
three major clades, with each clade run as a separate analysis in BEAST
v2.6.3 (see Dryad file S54).
Substitution rates were calculated directly in BEAUTi using the entire
500 SNP dataset. Coalescence rate was set at 10.0 and specified to
sample. Priors for both lambda and theta were set using a gamma
distribution. Lambda was set to have a distribution of G(2, 500). Theta
was set to have a distribution of G(1, 250). The chain length was set to
1,000,000, sampling every 1,000 generations. We utilized the BFD*
methods outlined by Leaché et al. (2014). Marginal likelihood values
were obtained using a path-sampling analysis in BEAST with 48 steps,
chain length of 1,000,000 generations and 10% burn-in. Resulting log
files were analyzed in Tracer v.1.7.1 (Rambaut and Drummond, 2005) to
assess ESS values and convergence of runs. Each model was run four times
and the marginal likelihood value was averaged across each clade in
order to further ensure convergence.