The power of relative rates tests depends on the data

Lindell Bromham*, David Penny, Andrew Rambaut, Michael D. Hendy

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

114 Citations (Scopus)

Abstract

One of the most useful features of molecular phylogenetic analyses is the potential for estimating dates of divergence of evolutionary lineages from the DNA of extant species. But lineage-specific variation in rate of molecular evolution complicates molecular dating, because a calibration rate estimated from one lineage may not be an accurate representation of the rate in other lineages. Many molecular dating studies use a 'clock test' to identify and exclude sequences that vary in rate between lineages. However, these clock tests should not be relied upon without a critical examination of their effectiveness at removing rate variable sequences from any given data set, particularly with regard to the sequence length and number of variable sites. As an illustration of this problem we present a power test of a frequently employed triplet relative rates test. We conclude that (1) relative rates tests are unlikely to detect moderate levels of lineage- specific rate variation (where one lineage has a rate of molecular evolution 1.5 to 4.0 times the other) for most commonly used sequences in molecular dating analyses, and (2) this lack of power is likely to result in substantial error in the estimation of dates of divergence. As an example, we show that the well-studied rate difference between murid rodents and great apes will not be detected for many of the sequences used to date the divergence between these two lineages and that this failure to detect rate variation is likely to result in consistent overestimation the date of the rodent-primate split.

Original languageEnglish
Pages (from-to)296-301
Number of pages6
JournalJournal of Molecular Evolution
Volume50
Issue number3
DOIs
Publication statusPublished - 2000
Externally publishedYes

Fingerprint

Dive into the research topics of 'The power of relative rates tests depends on the data'. Together they form a unique fingerprint.

Cite this