Skip to main content

Table 1 Model comparison using the harmonic mean estimator (HME) for each of the three data sets

From: Make the most of your samples: Bayes factor estimators for high-dimensional models of sequence evolution

Data set

Model

Root order

HME

sHME

log BF

Laurasiatheria

GTR

Independence (Second)

-173720.23

-173342.71

-

 

HG04

Second

-170093.65

-169819.78

3522.93

Pseudogenes

GTR

Independence (First)

-14026.11

-14001.45

-

 

HG04

First

-13772.80

-13738.83

262.62

Nuclear SSU rRNA

GTR

Independence (Zero)

-8441.69

-8409.53

-

 

HG04

Zero

-8412.27

-8374.33

35.20

 

GTR

Independence (First)

-8435.35

-8409.82

-

 

HG04

First

-8395.11

-8364.34

45.48

 

GTR

Independence (Second)

-8445.00

-8408.39

-

 

HG04

Second

-8399.92

-8353.75

54.64

  1. Harmonic mean estimates (HME) and stabilized harmonic mean estimates (sHME) for both the site-independent general time-reversible model (GTR) and the context-dependent model of Hwang and Green (HG04) [23], combined with the one or more ancestral root distributions [24]. Great care needs to be taken to make sure that under the GTR model, the same sites are taken into account when calculating the likelihood, this due to the presence of gaps. We emphasize this by stating which ancestral root distribution is used for the GTR model, on top of eliminating those sites for which the dependency pattern contains one or more gaps. The proposed context-dependent model (HG04) clearly outperforms the site-independent GTR model for all the data sets, according to the HME and sHME. Further, for the nuclear SSU rRNA data set and according to the sHME, the difference in model fit increases in favor of the HG04 model with increasing dependencies at the ancestral root.