Each simulation was based on a user-specified uniform prior
distribution of parameter ranges. The settings for the prior distributions
were empirically obtained after preliminary runs using
higher upper bounds. The obtained settings were q (0–30), m
(0–4 or 0–0 for the model assuming m = 0), and t (0–10). The
posterior distribution of parameter t included a peak at low
values with a flat tail over extended ranges of higher parameter
values. Thus, we chose a prior distribution with an upper bound
that did not include the flat tail of the distributions. The upper
bound of 10 for t was justified conservatively, as the estimate
of the tmrca (time of most recent common ancestor) parameter,
which is expected to be larger than t, was 5.267 (3.17–7.19,
90% Highest Posterior Density, HPD). Each run included a burnin
of one million steps followed by a run of a minimum of
10 million steps, with a minimum of 100,000 trees sampled using
a single chain. We ran three replicate runs for each analysis with
different random number starting seeds to ensure independent
runs gave similar results.
Each simulation was based on a user-specified uniform prior
distribution of parameter ranges. The settings for the prior distributions
were empirically obtained after preliminary runs using
higher upper bounds. The obtained settings were q (0–30), m
(0–4 or 0–0 for the model assuming m = 0), and t (0–10). The
posterior distribution of parameter t included a peak at low
values with a flat tail over extended ranges of higher parameter
values. Thus, we chose a prior distribution with an upper bound
that did not include the flat tail of the distributions. The upper
bound of 10 for t was justified conservatively, as the estimate
of the tmrca (time of most recent common ancestor) parameter,
which is expected to be larger than t, was 5.267 (3.17–7.19,
90% Highest Posterior Density, HPD). Each run included a burnin
of one million steps followed by a run of a minimum of
10 million steps, with a minimum of 100,000 trees sampled using
a single chain. We ran three replicate runs for each analysis with
different random number starting seeds to ensure independent
runs gave similar results.
การแปล กรุณารอสักครู่..
