We also discuss potential ways for optimizing phylodynamic analysis settings and developing more efficient algorithms for traversing the tree space. If interested, please check our preprint out! n/n
(More tree space animations of each dataset we analyzed are available at tinyurl.com/ssstreeanim.)

Nonetheless, some global parameters governed by general tree shape are rather robust to the sampling problems. E.g., the demographic trajectories are barely distinguishable among the chains, meaning that relatively short chains would be sufficient if such parameter is of primary interest. 6/

Tree space ruggedness and consequent sampling problems can strongly distort biological conclusions. E.g., our analyses of an HIV dataset reveal distinct early spread history of HIV among the chains. 5/
(The number of dispersal events between two focal geographic areas; each panel shows a chain.)

Our new MCMC diagnostics reveal that the sampling problems typically stem from a small part of the tree. E.g., in a LASV dataset, removing just two of the 551 sequences reshapes the tree space and effectively resolves the sampling problems. 4/
(Tree space samplers before and after the removal.)

We are interested in understanding phylodynamic tree landscape and its biological impacts. Here we show that the tree landscape is usually highly rugged, leading to widespread sampling problems. 3/
(MCMC progress in MDS tree space; each panel shows a dataset and each dot represents a tree.)

Unlike typical phylogenetic settings where the tree topology is our focus and the best tree is strongly supported by the informative genomic data, phylodynamic inference uses viral sequences with limited genetic diversity and frequently treats the topology as a nuisance. 2/