Title: On the Simplicity of Evolution Algorithms for Phylogenetic Networks
1On the Simplicity of Evolution Algorithms for
Phylogenetic Networks
Leo van Iersel and Steven Kelk
Part of this research has been funded by the
Dutch BSIK/BRICKS project AFM2.
2- Combine a set of small trees (triplets) into a
single network that is as simple as possible
algorithm
3Phylogenetic Trees
Root (common ancestor)
Time
Split vertices (ancestors)
Leaves (species)
4Phylogenetic Networks
Root
Split vertex
Reticulation vertex
Leaves
5Phylogenetic Networks
Root
- Reticulation can model
- Recombination
- Hybridization
- Horizontal gene transfer
- Ambiguity
Split vertex
Reticulation vertex
Leaves
6How Simple is a Network?
- Number of reticulations total number of
reticulation vertices (indegree two vertices) - Level maximum number of reticulation vertices in
a biconnected component
7Example level-2 network with 4 reticulations
blue biconnected component red reticulation
vertex
8New Algorithms
- Given dense set of triplets
- Construct level-1 network with a minimum number
of reticulation vertices - Given dense set of triplets
- Construct level-2 network with a minimum number
of reticulation vertices
9Triplets
abc
bca
acb
- A triplet set is dense if for each combination of
three leaves it contains at least one of the
three possible triplets
10Triplet Consistency
A triplet abc is consistent with a network if
this network contains a subdivision of abc
abc
11Triplet Consistency
Also triplet bca is consistent with this network
because it also contains a subdivision of bca
bca
12Previous results
- Aho et al. (1981)
- Constructing (phylogenetic) trees from triplets
in polynomial time - Jansson, Nguyen and Sung (2004)
- Constructing level-1 networks from dense triplet
sets in polynomial time - Van Iersel, Keijsper, Kelk, Stougie, Hagen and
Boekhout (2008) - Constructing level-2 networks from dense triplet
sets in polynomial time
13New Algorithms
- Given dense set of triplets
- Construct level-1 network consistent with the
input triplets that contains a minimum number of
reticulation vertices - Given dense set of triplets
- Construct level-2 network consistent with the
input triplets that contains a minimum number of
reticulation vertices
14Old algorithm
New algorithm
15SN-sets
- Definition. Subset S of the leaves is an SN-set
if there is no triplet xyz withx?S and y,z?S - Lemma Jansson and Sung.Any two SN-sets are
either disjoint or one is included in the other
16Observation. A set of leaves below a cut-arc is
an SN-set
Because then xyz with x?S and y,z?S is not
consistent with the network
17Algorithm for Level-1
- For each SN-set S, from small to large
- Construct a network consisting of a root
connected to two optimal networks NS1, NS2 for
included SN-sets S1, S2 - And all possible networks consisting of a cycle
connected to at least three optimal networks for
included SN-sets - The optimal solution NS is the network with the
minimum number of reticulation vertices over all
constructed networks
18Case 1 root is not in a cycle
S1 and S2 are the maximal SN-sets
19Case 2 root is in a cycle
Which SN-sets are S1, S2, S3, S4, S5?
20The Maximal SN-sets
Some maximal SN-sets may be divided below a path
ending in a reticulation vertex
21- Guess X
- Find maximal SN-sets that do not contain X
- There can still be a maximal SN-set below a path,
but
22- There exists a different solution where this
SN-set is below a single cut-arc - This solution has the same number of reticulations
23To conclude
- We know how to find the sets S1,,S5
- A single cycle can be constructed with the
algorithm by Jansson, Nguyen and Sung - Optimal networks NS1,,NS5 have been computed in
earlier iterations
24(No Transcript)
25Real network
Constructed network
26Conclusion and Discussion
- We can construct level-1 and level-2 phylogenetic
networks with a minimum number of reticulations
in polynomial time - Input triplet set has to be dense
- There must exist a network consistent with all
input triplets - Level can be at most two
27Thank you