Fast and Accurate Estimates of Divergence Times from Big Data

Abstract

Ongoing advances in sequencing technology have led to an explosive expansion in the molecular data available for building increasingly larger and more comprehensive timetrees. However, Bayesian relaxed-clock approaches frequently used to infer these timetrees impose a large computational burden and discourage critical assessment of the robustness of inferred times to model assumptions, influence of calibrations, and selection of optimal data subsets. We analyzed eight large, recently published, empirical datasets to compare time estimates produced by RelTime (a non-Bayesian method) with those reported by using Bayesian approaches. We find that RelTime estimates are very similar to Bayesian approaches, yet RelTime requires orders of magnitude less computational time. This means that the use of RelTime will enable greater rigor in molecular dating, because faster computational speeds encourage more extensive testing of the robustness of inferred timetrees to prior assumptions (models and calibrations) and data subsets. Thus, RelTime provides a reliable and computationally thrifty approach for dating the tree of life using large-scale molecular datasets.

Tempo and Mode of Genome Evolution in the Budding Yeast Subphylum

“…We used the same ML tree and calibrations as we did for the MCMCTree analysis (see above). Comparison of divergence time estimates between RelTime and MCMCTree revealed that they were broadly consistent (Pearson’s correlation coefficient r = 0.87, P-value < 2.2e-16; average time deviation = ~ 19.5%), which is in agreement with a previously published comparison of the two methods based on analyses of 8 empirical phylogenomic data matrices (Mello et al, 2017). Furthermore, both our study and that of Mello and colleagues (Mello et al, 2017) found that the RelTime estimates were generally older than MCMCTree estimates for the deep internodes of the budding yeast phylogeny (e.g., for the internodes between 12 major clades) and were generally younger than the MCMCTree estimates for shallower internodes (e.g., within the families Pichiaceae, Saccharomycodaceae, Saccharomycetaceae, Phaffomycetaceae, the CUG-Ala clade, and the CUG-Ser1 clade).…”

“…As RelTime is computationally much less demanding than MCMCTree (Mello et al, 2017), we conducted divergence time estimation using the complete 2408OG data matrix. We used the same ML tree and calibrations as we did for the MCMCTree analysis (see above).…”

Reliable confidence intervals for RelTime estimates of evolutionary divergence times

“…Currently, Bayesian HPD intervals are considered reliable estimates of uncertainties surrounding divergence time estimates, although they are not always the same as the 95% confidence intervals (CIs) in the frequentist statistics (Jaynes and Kempthorne 1976;MacKenzie et al 2017). Unfortunately, the enormous computational burden imposed by Bayesian approaches has hindered their applications to analyze many phylogenomic datasets (Pyron 2014;Mello et al 2017;Li et al 2019).…”

“…Non-Bayesian methods are currently limited to the use of minimum boundaries only, maximum boundaries only, or minimum and maximum boundary pairs as calibration constraints (Sanderson 2003;Tamura et al 2013), while Bayesian methods allow the usage of probability densities as calibrations and automatically accommodate interactions among them (Inoue et al 2010;Ho and Duchêne 2014). While Mello et al (2017) presented a simple procedure to derive minimum and maximum boundaries from the density distributions, this strategy does not consider interactions among calibrations and may lead to overestimates of the variance of divergence times (see below).…”

RelTime Rates Collapse to a Strict Clock When Estimating the Timeline of Animal Diversification

“…Advocates of RelTime claim that by eschewing fossils while estimating rates, RelTime avoids the negative impact of “flawed” calibrations in divergence time estimation (Tamura et al 2012; Battistuzzi et al 2015; Kumar and Hedges 2016). In a series of recent studies, RelTime has been benchmarked against Bayesian divergence time analyses, recovering comparable results in a fraction of the time (Mello et al 2017). However, in reproducing analyses of the timing of the animal diversification using the dataset of Erwin et al (2011), Battistuzzi et al (2015) recovered a much older Mesoproterozoic estimate for the origin of animals, akin to the results from early studies that relied on strict clock methods (Runnegar 1982; Wray et al 1996).…”