Application of Phylogenetic Networks in Evolutionary Studies

1
Application of Phylogenetic Networks in
Evolutionary Studies

• Daniel H. Huson and David Bryant
• Presented by Peggy Wang

2
The plan

• Terminology
• Split networks
• What are they?
• How can they be interpreted?
• Phylogenetic inference
• SplitsTree4

3
A tree of terms..
4
Terminology

• phylogenetic network
• any network in which taxa are represented by
  nodes and their evolutionary relationships are
  represented by edges

5
Types of networks

• phylogenetic tree
• Leaf labeled tree that represents the
  evolutionary history of a set of taxa, possibly
  with branch(edge) lengths, either unrooted or
  rooted.
• reticulate network
• Phylogenetic tree additional edges. Nodes with
  more than two parents represent reticulate events
  such as hybridization, recombination and hgt
• split network
• Represents incompatible and ambiguous signals in
  a data set. Parallel edges represents splits
  computed from data. Incompatible splits may
  result in nodes that do not represent ancestral
  species.

6
Wait whats a split again?

• split
• A partition of the taxa into two nonempty
  subsets, such as the partition obtained when we
  remove a branch from a phylogenetic tree.
• split network (formally)
• For a given taxon set X and set of splits S, we
  define a split network N to be a connected graph
  in which some of the nodes are labeled by taxa
  and all edges are labeled by splits.

7

• Removing all edges associated with a given split
  s in S divides N into two connected components,
  one part containing all taxa on one side of S and
  the other part containing all taxa on the other
  side.
• The edges along any shortest path in N are all
  associated with different splits.
• A split network contains exactly the same
  information as a list of splits with a weight for
  each split.

8

• Every split network represents a unique
  collection of splits.
• A given collection of splits can have many
  different split network representations.
• The interpretation of the network depends on how
  the splits were constructed and assigned weights

9
Interpreting Split Networks Representing
multiple trees

• We can use split networks to summarize a large
  collection of trees.
• Code each individual tree as a collection of
  splits
• Define a summary set of splits
• Represent the set using a split network.
• Consensus networks
• Constructed from all splits appearing in at least
  some fixed proportion of input trees

10
Interpreting Split NetworksRepresenting
multiple trees

• Confidence sets
• Assign an interval for the weights of each split
• A tree is contained within the split network N if
  
• Every split in the tree is a split in the network
• For every split in the tree, the corresponding
  branch length is contained within the
  corresponding interval
• For every split in the network not in the tree,
  the assigned interval contains zero.
• .

N
11
Interpreting Split NetworksRepresenting
multiple trees
Geometric interpretation Index splits from 1 to
m. Tree can be coded as a point in m-dimensional
space the ith coordinate is the length of the
ith split, or 0 if that split is not present in
the tree.
12
Interpreting Splits NetworksNetworks and
systematic error

• Sampling error
• Random error resulting from a small sample size
  (number of sites).
• Deal with these errors using nonparametric
  bootstrap, multiple samples from posterior
  distribution
• Systematic error
• Mistakes in the assumptions of a model or method
  which cause data to be misinterpreted. Likely to
  occur with large, multigene, heterogeneous data
  sets.
• How to deal with these errors?

13
Interpreting Splits NetworksNetworks and
systematic error

• Phylogenetic inference
• Construct a split network using the best
  available model and method.
• Determine if the network is significantly
  different from a tree.
• If the tree is significantly non-treelike, then
  there is probably an error in the model. If
  possible, improve the model and try again.
• If the network is treelike, and there is no
  significant sampling error, the continue with a
  tree-based phylogenetic analysis.

14
Reticulate Networks
2 disagreeing trees
Split network represents all splits present in
either of the two trees
Reticulate network Explains the differences in
the two trees using 3 reticulation events
15
SplitsTree4

• Integrates a wide range of phylogenetic network
  and phylogenetic tree methods, inference tools,
  data management utilities, and validation
  methods.
• Included methods for inferring split networks
• From character data. Median networks, parsimony
  splits, spectral analysis
• From distance matrices. Split decomposition and
  neighbor-net
• From sets of trees. Consensus networks and
  supernetworks.
• Also constructs other types of phylogenetic
  networks, eg recombination and hybridization
  networks
• User friendly?!

16
Example 1Heterogeneous Evolution
Jukes-Cantor p0.75 q0.05 0ltrlt0.4
17
Example 2Animal Phylogeny
Coelomate hypothesis Ecdysozoa hypothesis
18
More examples..Dusky dolphins
60 variables (sites of DNA) 35 haplotypes
Neighbor-joining tree with bootstrap values
Consensus network of 3 MP trees
Split decomposition network
95 confidence network
Median network
Neighbor-net network
19
Conclusion!

• Split networks are useful for visualization.
• However they are not useful for making conclusive
  phylogenetic analysis.
• SplitsTree4 encompasses many tools, but are they
  really that useful?