Bayesian tip-dating with continuous characters using BEASTmasteR

Abstract

Bayesian "tip-dating" is a method for simultaneously inferring the topology and dating of a phylogeny by including fossils as dated tips and conducting a total-evidence analysis, for example in the program BEAST2. Tip-dating is being actively explored, but all work to date has only used discrete (qualitative) characters for fossil tips. However, continuous (quantitative) characters are often available for fossil specimens, and ideally tip-dating analyses would include these as well. To enable such analyses, continuous-data capabilities were added to the R package BEASTmasteR. BEASTmasteR takes a data table of continuous traits and converts them into BEAST2 XML format, adding these characters to any DNA, amino acid, and/or discrete morphological data that is available. The likelihood of continuous traits on the tree is calculated using the Brownian motion model available in BEAST2 through modification of BEAST2's 2-dimensional continuous phylogeography model into a 1-dimensional model for any trait. Each trait is given a separate rate parameter which is also estimated. To test the validity of the model, continuous characters were simulated on an assumed tree (derived from a dated canid tree) with 22 tips (both fossil and living) under a Brownian motion model. Sets of 10, 25, or 100 continuous characters were generated and BEAST2 XML files were constructed using BEASTmasteR. Each BEAST2 inference was run for 50,000,000 generations. Inference on 10- or 25-character datasets converged quickly on the true tree, with the 25-character dataset showing higher posterior probabilities for many clades (only 4 branches with <50% posterior probability, PP) than the 10-character dataset (10 branches with <50% PP). Dating uncertainty also decreased by about 30%. However, the 100-character dataset failed to converge, perhaps because of the difficulty of jointly searching tree space and 100 rate parameters. The implications for practical analysis will be discussed, including the importance of the assumptions of independence between characters and independent rates. BEASTmasteR performance indicates that it should be helpful to researchers exploring continuous data: BEAST2 XML setup with continuous data takes <1 minute in BEASTmasteR, but at least 4 hours for an experienced user constructing the XML input by hand.