Title: Phylogenetic Tree Construction Methods and Programs
1Chapter 11 Phylogenetic Tree Construction
Methods and Programs
2- Character based methods
- Molecular sequences from individual taxa
- Characters at corresponding positions in
alignments are homologous - Characters of common ancestor can be traced
- Each character evolved independently
- Distance based methods
- Dissimilarity between sequence pairs, based on
alignment - All sequences involved are homologous
- Tree branches are additive
3Distance methods Clustering-based methods
- Unweighted Pair Group Method Using Arithmetic
Average (UPGMA) - Given distance matrix
- Group two taxa with smallest pairwise distance
- Generate new matrix with combined taxa as a
single group - Repeat grouping and reduce matrix again
- Continue iteration until all taxa are grouped
- Last taxon added considered outgroup and tree
root
4(No Transcript)
5Distance methods Clustering-based methods
Neighbour Joining Similar to UPGMA, but does not
assume constant evolutionary rate between
taxa Makes use of distance matrix, but corrects
for unequal evolutionary rates by calculating
r-distances and transformed r-distances dAB
dAB -1/2(rArB) dAB is the converted distance
between A and B and dAB is the actual
evolutionary distance between A and B rA (or rB)
is sum of distances of A (or B) to all other
taxa ri ?dij Where i and j are two different
taxa r value is needed to create a modified
distance matrix Transformed r ri
ri/(n-2) Used to determine distance of
individual taxon to nearest node
6C
7(No Transcript)
8- Generalized neighbour joining
- Disadvantage of NJ is that one tree is generated
- Depending on choice of two equally close starting
taxa, sub-optimal tree may be calculated - Generalized NJ multiple NJ runs with different
starting taxa - Select tree from all calculated trees that best
fit actual evolutionary distances
9- Optimally based methods Fitch-Margoliash
- Selects best tree based on minimal deviation
between calculated distance and original distance - Randomly selects two taxa, and calculated branch
lengths - Examines all tree topologies and chooses one with
minimum deviation - dij pairwise distance
- pij corresponding tree branch length
E
10Minimum evolution (ME) Similar to FM, but uses
different optimality criterion to find tree with
minimum branch length S ?bi bi is ith branch
length ME slightly outperforms FM
11Character based Methods Maximum Parsimony
(thriftiness) Based on character differences in
sequence alignments Number of differences
counted, i.e., ancestral sequences can be
inferred Tree with fewest evolutionary changes
(Occams Razor) Occams Razor (14th-century
English logician and Franciscan friar, William of
Ockham.) "entia non sunt multiplicanda praeter
necessitatem", roughly translated as "entities
must not be multiplied beyond necessity, i.e.,
"All other things being equal, the simplest
solution is the best."
12William of Ockham - Sketch labelled "frater
Occham iste", from a manuscript of Ockham's Summa
Logicae, 1341
13- Maximum Parsimony
- Search all possible tree topologies
- Reconstruct ancestral tree that require minimum
number of character changes - Use only sites with rich phylogenetic information
to save time - Informative sites two different kinds of
characters occurring at least twice
14Find tree with minimum number of character changes
15- Weighted parsimony
- Weighing scheme takes into account functionally
important sites - Transitions versus transversion
16- Tree searching methods
- Only works for lt 10 taxa
- Need simplification steps for more taxa branch
and bound
17- Branch and Bound
- Upper limit for number of allowed sequence
variations - Build tree for all taxa involved using UPGMA or
NJ - Compute minimum substitutions for such a tree
- MP must be equal or better than UPGMA or NJ
18- When number of taxa gt 20
- Branch and bound computationally too intensive
- Need heuristic approach
- Cut of branch and re-graft
- Recalculate branch lengths
- Continue iteration until no shorter branch
arrangements are found - Can get stuck in local minima
- Global sub-pruning option in some programs
19- Long Branch Attraction
- Rapidly evolving taxa with long branches are
places together in a tree - Assumption that all taxa evolve at the same rate
- All mutations (transitions and transversions)
contribute equally to branch lengths - Weighted parsimony should solve problem
- Increase taxon sampling size
20- Maximum likelihood method
- The probability that a nucleotide changes over
time t is calculated from - Pt 1/4 1/4e-?t
- The probabilities are multiplied from root to tip
for all branches - All possible tree topologies are examined, and
probabilities calculated - The tree with the highest probability is chosen
21- Quartet Puzzling
- Make many subsets composed of four taxa
- Combine all quartets into single tree
- Significantly reduces computing time
- ML tree not guaranteed
-
22- Neighbour Joining Maximum Likelihood
- Construct initial tree with NJ
- Poor branches (low bootstrap support),
collapsed - Resolve by ML
- 10X faster than neat ML method
23- Genetic Algorithm
- Generate random population of trees with
arbitrary branch lengths - Select tree with highest likelihood score
- Allow this tree to mutate
- Screen mutated offspring, and select highest
likelihood trees - Repeat process until no higher scoring trees are
produced
24Bayesian Analysis Based on posterior
probability Probably the fastest and most
accurate phylogenetic analysis available to date
Posterior probability
25- Phylogenetic Tree Evaluation
- Bootstrapping
- Recalculate trees with randomly perturbed
datasets - If the original tree topology is repeatedly
found, it is most likely correct - Parametric
- New datasets are generated based on sequence
distribution - Based opn sequence substitution model
(Jukes-Cantor or Kimura) - Non-parametric bootstrapping
- Random replacement of sites
- Jack knifing
- Randomly delete half of the sites in a dataset
26(No Transcript)
27- Bayesian simulation
- Most efficient in terms of statistical evaluation
- Does not require bootstrapping because MCMC
procedure millions of re-sampling steps - Posterior probabilities assigned as statistical
support - Makes statistical valuation of ML trees more
feasible
28- Kishino-Hasegawa Test
- Test whether two competing tree topologies are
statistically significantly different - Difference in branch length at each informative
site is calculated - Standard deviation of the difference values is
them calculated - Allowing derivation of t-value
29Shimodaira-Hasegawa Test Frequently used for
ML Based on ?2 test Converted to p-value
30Phylogenetic Programs PAUP http//paup.csit.fsu.
edu/ Phylip http//evolution.genetics.washington.e
du/phylip.html TREE-PUZZLE http//www.tree-puzzle.
de PHYML http//atgc.lirmm.fr/phyml/ MrBayes
http//morphbank.ebc.uu.se/mrbayes
31(No Transcript)