Skip to main content

Table 2 Simulated data: comparative performance under correct and incorrect models. Performance of maximum-likelihood and Bayesian phylogenetic inference without, and with, violation of the model of protein sequence change, for trees with one, or two, relatively long branches.

From: Bayesian and maximum likelihood phylogenetic analyses of protein sequence data under relative branch-length differences and model violation

One long branch

 

No model violation 1

Model violation 1

 

First2

Wrong

Burd SD

Mean SD

Burd ED

Mean ED

First

Wrong

Burd SD

Mean SD

Burd ED

Mean ED

ML3

30

56

144

2.57

56

1.00

20

65

164

2.52

65

1.00

BUU

30

46

232

5.04

46

1.00

20

50

222

4.44

50

1.00

BGU

30

36

96

2.67

36

1.00

20

39

118

3.03

39

1.00

BGE

30

28

88

3.14

28

1.00

20

39

116

2.97

39

1.00

Two long branches

 

No model violation

Model violation

 

First

Wrong

Burd SD

Mean SD

Burd ED

Mean ED

First

Wrong

Burd SD

Mean SD

Burd ED

Mean ED

ML

20

186

1124

6.04

273

1.47

20

174

1166

6.70

207

1.19

BUU

20

237

1854

7.82

326

1.38

10

244

1900

7.79

299

1.23

BGU

20

87

270

3.10

104

1.20

20

105

468

4.46

119

1.13

BGE

20

86

314

3.65

101

1.17

20

115

650

5.65

131

1.14

  1. 1 Protein-sequence data were evolved under the Jones et al. (JTT) or, alternatively, mammalian mitochondrial (mtmam) model of sequence change, and trees were inferred assuming the JTT model.
  2. 2 Performance was measured by six indices: First, the lowest investigated branch-length ratio at which at least one inaccurately reconstructed tree was found; Wrong, the number of inaccurately inferred trees out 400 (8 branch-length ratios × 50 replicates at each ratio); BurdSD, the bipartition burden, calculated as the Robinson-Foulds symmetric distance by which each tree differs from the known tree, summed over the 400 trees; MeanSD, the mean Robinson-Foulds symmetric distance per inaccurate tree; BurdED, the edit burden, calculated as the edit distance by which each tree differs from the known tree, summed over the 400 trees; and MeanED, the mean edit distance per inaccurate tree.
  3. 3 Inference methods: ML, protein maximum likelihood with gamma ASRV correction; BUU, Bayesian inference, uncorrected for ASRV, uniform prior; BGU, Bayesian inference, gamma ASRV correction, uniform prior; and BGE, Bayesian inference, gamma ASRV correction, exponential prior. See text for further details.