Phylogenetic targeting - Bioinformatics Leipzig

Download Report

Transcript Phylogenetic targeting - Bioinformatics Leipzig

Three Weeks of Experience at the
formatics Institute
Christian Arnold
Bioinformatics Group, University of Leipzig
Bioinformatics Herbstseminar
October 23th, 2009
Content
1. The 10kTrees Project
2. Phylogenetic Targeting
3. Acknowledgements
1.The 10kTrees Project
Goals
• Updated primate phylogeny that includes phylogenetic
uncertainty
– Use newest available sequence data, include as much primate
species as possible, and update regularly
– Produce a set of >=10,000 primate-wide trees (with branch
lengths) that are appropriate for taxonomically broad
comparative research on primate behavior, ecology and
morphology using Bayesian methods
• Make it accessible to other researchers
Methodology
1. Download
sequences from
GenBank
2. Select the longest
available sequence
for each gene in each
species
3. Create individual
fasta file with all
available sequences
for each gene
4. Create availability
matrix
5. Identify species
with non-overlapping
genes
6. Create MSA using
Muscle
7. Improve alignment
quality using
GBLOCKS
8. Identify best
substitution model for
each gene
9. Concatenate
sequences and create
partitioned dataset in
MrBayes format
10. Run MrBayes
11. Evaluate MrBayes
analysis and calculate
consensus tree
12. Update website
Version 1 vs. Version 2
Version 1
Version 2
Species
187
231
Genes
4 mitochondrial (COI, COII,
CYTB and ND1) and 1
autosomal gene (SRY)
6 mitochondrial (12S rRNA,
16S rRNA, COI, COII, CYTB,
cluster of other mitochondrial
genes) and 3 autosomal genes
(SRY, CCR5, MC1R)
Genetic loci
2
4
Total No. of Sites
5134
~9000
Collected sequences
413 out of 935 total
(55.8% missing data)
1007 out of 2079 total
(51.6% missing data)
No. of constraints
29
1
Generations
8 millions
60 millions
Computing time
~ 48 days (16 processors in
parallel, ~ 3 days each)
~ 2 years (32 processors in
parallel, ~ 3 weeks each)
Preliminary consensus tree
Green: Cercopithecines
Blue: Hominoids
Red: Platyrrhines
Yellow: Tarsiers
Brown: Strepsirrhines
Rooted with Galeopterus
variegatus
The 10kTrees Website
http://10ktrees.fas.harvard.edu/
Current Progress
• Submitted to Evolutionary Anthropology, in press.
• Will be presented at the AAPA conference (April 2010)
in Albuquerque, New Mexico
• Version 2 is almost finished
• Available at http://10kTrees.fas.harvard.edu
Summary
• Bayesian approach is time-consuming, but works well, even
though data matrix is very sparse
• Increased number of sequences in Version 2 dramatically
reduces need for constraints and improves quality of tree and
branch lengths estimates
• Ongoing project
• Total number of downloaded trees since June 2009: 95800
2. Phylogenetic Targeting
Which species should we study?
Goals
For which species should we collect data in order to increase
the size of comparative data sets ?
Example 1/2
• Hypothesis: Two characters (x and y) show correlated evolution
• Goal: Test this hypothesis comparatively (e.g. by using
phylogenetically independent contrasts and correlation tests)
• Problem 1: Data has been only collected for x, but not for y
• Solution 1: Collect data for y and test hypothesis
• Problem 2: From which species should we collect data for y?
• Solution 2: Phylogenetic targeting!?
Example 2/2
Brain size Cognitive data
s1
4
?
s2
9
7
s3
10
?
s4
3
?
s5
2
?
Collecting new data is time-consuming and expensive…
Methods
• Systematically generate all possible pairwise comparisons
• For every pairwise comparison, calculate character differences for
the two species that form the pair and assign a score
• Determine set of phylogenetically independent pairs that
maximizes the sum of all selected pair scores (maximal pairing)
s1
s2
s3
s4
s5
s6
s7
Maximal pairing: Example
Decomposition of the maximal pairing
 S desc(T )
ST  max 
max R ( S R   S subtrees( R ) )
2
Time complexity: O(n 3 ) , for balanced trees: O(n log 2 n)
Simulation results 1/2
Detecting correlated character evolution, based on
selection of 12 species
• Random (Rnd) selection of species
– Type 1 errors close to nominal level
– Power: ~40%, independent of number of taxa
– Uses 67% of available variation
• Phylogenetic targeting (PT) induced selection of species
– Type 1 errors close to nominal level
– Power: 67-81%, increases with number of taxa
– Uses 89% of available variation
Fraction of available variation after sampling
Simulation results 2/2
PT
Rnd
12
PT
18
Rnd
PT Rnd
24
PT Rnd
Number of selected species
Current Progress
• A revised version will be resubmitted to American Naturalist
in the not too distant future
• TODO: Extend simulations and clarify some issues
• Available at http://phylotargeting.fas.harvard.edu
Summary
• A focused selection of species can save valuable time and
money
• Phylogenetic targeting provides a very flexible approach and
can address different questions in the context of limited
resources
• Dynamic programming algorithms are everywhere
3. Acknowledgements
Thanks!
• Harvard University
• Max-Planck Institute for
Evolutionary Anthropology
• University of Leipzig
• Charlie Nunn
• Luke Matthews
• Peter F. Stadler
Any Questions?
Thank you for your attention!
Questions?
If not: Cheers (it’s early, but not too
early…)