Abstract
There are a growing number of areas, e.g. epidemiology and within-organism cancer evolution, where re-analysing all available data from scratch every time new data becomes available or old data is refined is no longer feasible. All these and related areas can benefit from online phylogenetic inference that can booster previous data analyses.
Here, we make the case that adding/removing taxa from an analysis can have substantial non-local impact on the tree that is inferred, both in a model based setting, as well as for distance based methods. Consequently, online phylogenetic algorithms may require global updates of the trees and other parameters, a task that in general is highly non-trivial.
Motivated by this observation, we designed an online algorithm that benefits from a parallelism in a Bayesian setting that is substantially more efficient than re-running the analysis from scratch. Furthermore, our algorithm is not sensitive to the number of sequences added, allowing the sequence data to grow/be refined iteratively. We show how this approach can be used in a maximum likelihood setting, and – apart from adding/removing new sequences – demonstrate a number of practical alternative use cases of our algorithm, including how to break up a single (offline) large analysis to get results faster.
An open source implementation is available under GPL3 license as the ‘online’ package for BEAST 2 at https://github.com/rbouckaert/online and a tutorial at https://github.com/rbouckaert/online-tutorial.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
alex{at}biods.org.
AG and LC acknowledge support from the Royal Society Te Apārangi through a Rutherford Discovery Fellowship (RDF-UOO1702). This work was partially supported by Ministry of Business, Innovation, and Employment of New Zealand through an Endeavour Smart Ideas grant (UOOX1912) and a Data Science Programmes grant (UOAX1932).