Accessing the relative performance of fast molecular dating methods for phylogenomic data

Abstract

Due to advances in genome sequencing techniques, there was a significant growth of phylogenomic datasets. This massive amount of data represents a computational challenge for molecular dating with the Bayesian approach that relax the assumption of rate constancy. To overcome these issues, over the last few decades, rapid molecular dating methods have been proposed. However, comparative evaluation of their relative performances on empirical data sets is lacking. We analyzed 23 empirical phylogenomic datasets to investigate the performance of two commonly employed fast dating methodologies, the penalized likelihood (PL), implemented in treePL, and the relative rate framework (RRF), implemented in RelTime. They were compared to Bayesian analysis under the same models and calibration settings. We found that the RRF was computationally faster and generally provided node age estimates statistically equivalent to Bayesian divergence times. Furthermore, contrasted to Bayesian dating, PL time estimates were excessively precise. To approximate Bayesian approaches, RelTime is an efficient method with significantly lower computational demand, being up to more than 100 times faster than treePL. Thus, to alleviate the computational burden of Bayesian divergence time inference in the era of massive genomic data, molecular dating can be facilitated using the RRF, so that evolutionary hypotheses can be tested more quickly and efficiently.

Resolving evolutionary relationships in the groundsels: phylogenomics, divergence time estimates, and biogeography ofPackera(Asteraceae: Senecioneae)

“…Divergence time estimates between RelTime and treePL differed (i.e., ~6MY difference for crown Packera), with treePL producing overall older estimates than RelTime. This is not surprising given that previous studies testing the efficacy of Bayesian vs. non-Bayesian dating methods have shown that treePL typically overestimates the ages and has narrower confidence intervals than RelTime (Costa et al, 2022). Even so, treePL is still being used with larger datasets (i.e., Li et al, 2019;Janssens et al, 2020;Maurin et al, 2022) given it is computationally less intense than RelTime and other Bayesian dating methods (Mello et al, 2017;Tao et al, 2020;Barba-Montoya, 2021;Costa et al, 2022).…”

“…Estimates of divergence times were calculated on the best fitting nuclear tree using two relaxed-clock methods: treePL v. 1.0 (Smith & O'Meara, 2012) which utilizes the penalized likelihood (PL) method (Sanderson, 1997, 2002), and RelTime (Tamura et al, 2012, 2018 which utilizes the relative rate framework (RRF). Both methods are efficient for larger datasets, though recent studies have shown that RelTime produces more similar results to Bayesian approaches than treePL or other non-Bayesian dating methods (Tao et al, 2020;Barba-Montoya, 2021;Costa et al, 2022). Even so, both methods were tested to check for consistency in divergence time estimates.…”