Speaker
Description
Tree metrics are measures to compare the similarity between two tree topologies. Effective sample size (ESS) is a statistic that quantifies the amount of autocorrelation in Markov chain Monte Carlo (MCMC) and is used to assess run convergence. With tree metrics, Lanfear et al showed that we can compute an ESS for tree topologies. From the plethora of tree metrics that have been developed, which is best for estimating tree ESS? We evaluated tree ESS estimation using 13 different tree metrics. First, we computed the tree ESS for toy models where we could directly infer the true tree ESS and then we explored how tree ESS varies with a model that has an increasing number of random nearest neighbour interchange or subprune regraft moves between trees. Finally, we computed the tree ESS for MCMC runs of viral outbreaks including: SARS-COV2, RSV and HCV using the BEAST2 software. While we found that the tree metrics could accurately estimate the tree ESS for the toy models, there was considerable variation in the estimates of tree ESS for the random moves models and empirical viral models. Our results highlight that there is no superior tree metric for computing tree ESS and there may be room to develop a new tree metric for computing statistics like the tree ESS.
| Expedited Notification | No thanks, I do not require Expedited Notification |
|---|
