Skip to main content

Table 4 Comparison of molecular clock and demographic models for four empirical data sets: Enterovirus A71 (EV-A71), West Nile Virus (WNV), Rabbit Hemorrhagic Disease Virus (RHDV), and Shigella sonnei

From: Cross-validation to select Bayesian hierarchical models in phylogenetics

Method

Data set

SC + CSC

SC + EGC

UCLN + CSC

UCLN + EGC

Cross validation (50 % training; 50 % test)

EV-A71

−1129.4(±3.1)

−1122.3(±2.0)

−1921.9(±9.8)

−1396.1(±12.0)

WNV

−8216.7(±1.3)

−8213.1(±2.5)

−8648.9(±5.3)

−8691.3(±5.0)

RHDV

−6456.1(±0.6)

−6908.8(±0.3)

−6102.8(±1.3)

−6101.9(±1.6)

Shigella sonnei

−7698.8(±0.4)

−7699.5(±0.3)

−25997.4(±7.9)

−25630.9(±6.3)

Cross validation (80 % training; 20 % test)

EV-A71

−443.0(±1.8)

−440.8(±1.0)

−1246.5(±4.7)

−1286.7(±14.8)

WNV

−3615.2(±2.6)

−3614.9(±2.5)

−3900.0(±19.9)

−3857.1(±19.2)

RHDV

−2394.7(±0.6)

−2393.5(±0.7)

−2336.7(±1.0)

−2279.2(±0.8)

Shigella sonnei

−2978.2(±1.9)

−2979.8(±2.0)

−3172.3(±11.6)

−3032.5(±10.0)

Marginal likelihoods using stepping stone

EV-A71

−2017.0

−2014.7

−2017.9

−2078.6

WNV

−18012.7

−17998.2

−18009.2

−17991.4

RHDV

−11323.8

−11292.6

−11271.5

−11245.8

Shigella sonnei

−14739.6

−14746.5

−14717.8

−14717.8

  1. The models correspond to four combinations of clock and demographic models: strict clock (SC), uncorrelated lognormal clock (UCLN), constant-size coalescent (CSC), and exponential-growth coalescent (EGC). Mean log likelihoods across ten replicates are given for the test set from each data set, using training sets of 50 and 80 % of the total alignment length. Marginal log likelihoods using stepping-stone sampling are also shown for comparison. Values in bold correspond to the highest log likelihood in each case. Values in parentheses indicate the standard error around the mean likelihood for ten cross-validation replicates