Fast and slow implementations of relaxed-clock methods show similar patterns of accuracy in estimating divergence times

Abstract

Phylogenetic analyses are using increasingly larger data sets for estimating divergence times. With this increase in data sizes, the computation time required is becoming a bottleneck in evolutionary investigations. Our recent study of two relaxed-clock programs (BEAST and MultiDivTime [MDT]) showed their usefulness in time estimation; however, they place a significant computational time burden on biologists even for moderately small data sets. Here, we report speed and accuracy of another relaxed-clock program (MCMCTree, MC2T). We find it to be much faster than both MDT and BEAST while producing comparable time estimates. These results will encourage the analysis of larger data sets as well as the evaluation of the robustness of estimated times to changes in the model of evolutionary rates and clock calibrations.

FIG. 1.

Computational times of MDT, BEAST, and MC2T for a 14 sequence phylogeny (model-match case). (A) Phylogeny used to simulate sequences; calibration node pairs are h and k (46 and 10 Ma), h and d (46 and 81 Ma), and j and k (23 and 10 Ma). Distribution of computational times for (B) MDT and (C) BEAST. (D) Relationship between the number of sequences and the computational time for MDT and MC2T. (E) Overall, comparison of computational times on a log scale for MC2T, MDT, and BEAST.

FIG. 2.

Distributions of the normalized difference (%) of estimated times from true times for slow (MDT and BEAST; lines) and fast methods (MC2T; histograms) when (A, B) the correct lineage rate variation (model match) and (C, D) the incorrect model (model violation) were used. Graphs in the inset show histograms for the absolute values of the normalized differences. Numbers in parentheses are the mean and standard deviations of the distribution (in order). In the main panels, a negative value on the x axis shows the degree of underestimation, and positive values indicate overestimations.