Loading in 5 sec....

creativecommons/licenses/by-sa/2.0/PowerPoint Presentation

creativecommons/licenses/by-sa/2.0/

- 100 Views
- Uploaded on
- Presentation posted in: General

creativecommons/licenses/by-sa/2.0/

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

CIS 786, Lecture 2

Usman Roshan

- Study of how species relate to each other
- “Nothing in biology makes sense, except in the light of evolution”, Theodosius Dobzhansky, Am. Biol. Teacher (1973)
- Rich in computational problems
- Fundamental tool in comparative bioinformatics

- Study of evolution
- Origin and migration of humans
- Origin and spead of disease

- Many applications in comparative bioinformatics
- Sequence alignment
- Motif detection (phylogenetic motifs, evolutionary trace, phylogenetic footprinting)
- Correlated mutation (useful for structural contact prediction)
- Protein interaction
- Gene networks
- Vaccine devlopment
- And many more…

U

V

W

X

Y

AGGGCAT

TAGCCCA

TAGACTT

TGCACAA

TGCGCTT

X

U

Y

V

W

- Phylogenies are equivalent to bipartitions

- Two main methodologies:
- Alignment first and phylogeny second
- Construct alignment using one of the MANY alignment programs in the literature
- Do manual (eye) adjustments if necessary
- Apply a phylogeny reconstruction method
- Fast but biologically not realistic
- Phylogeny is highly dependent on accuracy of alignment (but so is the alignment on the phylogeny!)

- Simultaneously alignment and phylogeny reconstruction
- Output both an alignment and phylogeny
- Computationally much harder
- Biologically more realistic as insertions, deletions, and mutations occur during the evolutionary process

- Alignment first and phylogeny second

- Compute alignment (for now we assume we are given an alignment)
- Construct a phylogeny (two approaches)
- Distance-based methods
- Input: Distance matrix containing pairwise statistical estimation of aligned sequences
- Output: Phylogenetic tree
- Fast but less accurate

- Character-based methods
- Input: Sequence alignment
- Output: Phylogenetic tree
- Accurate but computationally very hard

- Poisson process
- Number of changes in a fixed time interval t is independent of changes in any other non-overlapping time interval u
- Number of changes in time interval t is proportional to the length of the interval
- No changes in time interval of length 0

- Let X be the number of nucleotide changes on a single edge. We assume X is a Poisson process
- Probability dictates that

- We want to compute (the probability of a nucleotide change on edge e)
- The probability of observing a change is just the sum of probabilities of observing k changes over all possible values of k (excluding even ones because those changes cannot be seen)

- Expected number of nucleotide changes on a given edge is given by
- Key: is additive

- Assume we have a path of k edges and that p1, p2,…, pk are the probabilities of change on each edge of the path
- Using induction we can show that
- Multiplicative term is hard to deal with and does not easily decompose into a product or sum of pi’s

- But the expected number of nucleotide changes on the path p is elegant

- Simple 0,1 alphabet evolutionary model
- i.i.d. model
- uniformly random root sequence

- Jukes-Cantor:
- Uniformly random root sequence
- i.i.d. model

- General Markov Model
- Uniformly random root sequence
- i.i.d. model
- For time reversible models

- Standard assumption of how sites can vary is that each site has a multiplicative scaling factor
- Typically these scaling factors are drawn from a Gamma distribution (or Gamma plus invariant)

- Molecular clock: the expected number of changes for a site is proportional to time
- No-common-mechanism model: there is a random variable for every combination of edge and site

- For sequences A and B what is the evolutionary distance under the Jukes-Cantor model?
- ACCTGTGGGTAACCACCC
- ACCTGAGGGATAGGTCCG

- But we don’t know what is

- Assume nucleotide changes are Bernoulli trials (i.i.d. trials of success or failure)
- is probability of head in n Bernoulli trials (n is sequence length)
- Compute a maximum likelihood estimate for

- ACCTGTGGGTAACCACCC
- ACCTGAGGGATAGGTCCG
- 0 0 0 0 0 1 0 0 0 1 1 0 0 0 1 0 0 1

- We want to find the value of p that maximizes the probability:
- Set dP/dp to 0 and solve for p to get

- = 5/18
- Continuing in this manner we estimate for all pairs of sequences in the alignment
- We now have a distance matrix under a biologically sound evolutionary model

- ACCTGTGGGTAACCACCC
- ACCTGAGGGATAGGTCCG
- 0 0 0 0 0 1 0 0 0 1 1 0 0 0 1 0 0 1

- UPGMA: similar to hierarchical clustering but not additive
- Neighbor-joining: more sophisticated and additive
- What is additivity?

UPGMA is not additive but works for

ultrametric trees. Takes O(n^2) time

B

A

C

D

A

6

26

26

10

10

26

26

B

6

C

3

3

3

3

D

A

C

D

B

- Initialize n clusters where each cluster i contains the sequence i
- Find closest pair of clusters i, j, using distances in matrix D
- Make them neighbors in the tree by adding new node (ij), and set distance from (ij) to i and j as Dij/2
- Update distance matrix D: for all clusters k do the following (ni and nj are size of clusters i and j respectively)
- Delete columns and rows for i and j in D and add new ones corresponding to cluster (ij) with distances as computed above
- Goto step 2 until only one cluster is left

B

A

C

D

13

13

A

6

26

26

26

26

B

3

6

3

C

3

3

D

A

C

D

B

Doesn’t work (in general) for non-ultrametric

trees

B

A

C

D

3

3

A

13

16

26

3

3

12

19

B

10

10

B

C

13

C

D

D

A

UPGMA constructs incorrect tree here

7.25

B

A

C

D

7.25

A

13

16

26

7.25

7.25

12

19

B

6

6

13

C

B

A

D

C

D

Bipartition (BC,AD) is not in true tree

7.25

3

3

3

3

7.25

7.25

7.25

10

10

B

C

6

6

D

A

B

A

D

C

True tree

UPGMA tree

- Additive and O(n^2) time
- Initialization: same as UPGMA
- For each species compute
- Select i and j for which is minimum
- Make them neighbors in the tree by adding new node (ij), and set distance from (ij) to i and j as

- Update distance matrix D: for all clusters k do the following
- Delete columns and rows for i and j in D and add new ones corresponding to cluster (ij) with distances as computed above
- Go to 3 until two nodes/clusters are left

NJ constructs the correct tree for additive

matrices

B

A

C

D

3

3

A

13

16

26

3

3

12

19

B

10

10

B

C

13

C

D

D

A

- The true evolutionary tree is never known in practice. Simulation allows us to study accuracy of methods under biologically realistic scenarios
- Mathematics behind the phylogenetics is often complex and challenging. Simulation allows us to study algorithms when not possible theoretically and also examine algorithm performance under various conditions such as different evolutionary rates, sequence lengths, or numbers of taxa

- As sequence lengths tend to infinity the distance estimation improves and eventually leads to the true additive matrix
- If a method like NJ is then applied we get the true tree.
- In practice, however, we have limited sequence length. Therefore we want to know how much sequence length a method requires to achieve low error

Can be studied experimentally or theoretically

Theoretical results offer loose bounds

Experiments (under simulation) provide more realistic bounds on sequence lengths

Sequence lengths required to obtain 90% accuracy

- Later we will look at Disk-Covering Methods and study sequence length requirements of other methods (in addition to NJ)

- Character based method
- NP-hard (reduction to the Steiner tree problem)
- Widely-used in phylogenetics
- Slower than NJ but more accurate
- Faster than ML
- Assumes i.i.d.

- Input: Set S of n aligned sequences of length k
- Output: A phylogenetic tree T
- leaf-labeled by sequences in S
- additional sequences of length k labeling the internal nodes of T
such that is minimized.

- Input: Four sequences
- ACT
- ACA
- GTT
- GTA

- Question: which of the three trees has the best MP scores?

ACT

ACT

ACA

GTA

GTT

GTT

ACA

GTA

GTA

ACA

ACT

GTT

ACT

ACT

ACA

GTA

GTT

GTA

ACA

ACT

2

1

1

3

3

2

GTT

GTT

ACA

GTA

MP score = 7

MP score = 5

GTA

ACA

ACA

GTA

2

1

1

ACT

GTT

MP score = 4

Optimal MP tree

Optimal labeling can be

computed in linear time O(nk)

GTA

ACA

ACA

GTA

2

1

1

ACT

GTT

MP score = 4

Finding the optimal MP tree is NP-hard

Local optimum

Cost

Global optimum

Phylogenetic trees

- Determine a candidate solution s
- While s is not a local minimum
- Find a neighbor s’ of s such that MP(s’)<MP(s)
- If found set s=s’
- Else return s and exit

- Time complexity: unknown---could take forever or end quickly depending on starting tree and local move
- Need to specify how to construct starting tree and local move

- Random phylogeny---O(n) time
- Greedy-MP

Greedy-MP takes O(n^2k^2) time

For each edge we get two different topologies

Neighborhood size is 2n-6

Neighborhood size is quadratic in number of taxa

Computing the minimum number of SPR moves between two rooted phylogenies is NP-hard

- Neighborhood size is cubic in number of taxa
- Computing the minimum number of TBR moves between two rooted phylogenies is NP-hard

Local search

Local optimum

Local search

Local optimum

Perturbation

Output of perturbation

Local search

Local optimum

Perturbation

Local search

Output of perturbation

- Ratchet
- Iterative-DCM3
- TNT

- Performance studies on local search for MP
- Maximum likelihood
- Alignment