Below are tables and illustrations to show the quality of the method. Click on the thumbnails to get PostScript versions of the pictures.
Here is our estimate of the nucleotide substitution matrix for the six primates. The rows and columns are ordered as ACGT.
| -0.858 | 0.232 | 0.877 | 0.115 |
| 0.139 | -1.316 | 0.207 | 0.539 |
| 0.566 | 0.223 | -1.239 | 0.139 |
| 0.115 | 0.882 | 0.215 | -0.800 |
Yang (J. Mol. Evol. 39:105-111, 1994) found this estimate of the same matrix using a maximum likelihood method.
| -0.818 | 0.221 | 0.909 | 0.100 |
| 0.132 | -1.349 | 0.215 | 0.537 |
| 0.586 | -0.231 | -1.322 | 0.128 |
| 0.100 | 0.897 | 0.198 | -0.765 |
With these diagrams, we want to demonstrate that the distances we compute are linear, as is required by neighbor joining and other distance calculation methods.
|
|
|
|
The diagrams are plotting distances d1against d2, d1 against d3 and d2 against d3. The three distances have been produced by the three distance informative eigenvectors of the estimate of the nucleotide substitution matrix for the sequences. Each dot represents one sequences pair. The lines are found by a least squares fitting of the dots.
neighbor in the PHYLIP
package) from the distances distance computed from the six
primates. Using
fitch (also PHYLIP) yeilds basically the same tree, with
sum-of-squares deviation 0.00126, except
that it puts the human-chimp-gorilla branching differently. If
fitch is used on the output from
dnadist (PHYLIP) on the same data, the average
standard deviation is 0.00161 both when the Kimura distance
and the maximium likelihood distances (based on a Kimura model
with different nucleotide frequences) are used.
The image was produced using treetool.
| Entry page | Source code | Manual | Test data | Illustrations | Links |