Phylogenetic Shadowing

1
Phylogenetic Shadowing

• Daniel L. Ong

2
Abstract

• The human genome contains about 3 billion base
  pairs!
• Algorithms to analyze these sequences must be
  linear to be tractable
• Finding genes is important to Molecular
  Biologists, first step to understanding

3
Outline

• Introduction
• Alignments
• Phylogenetic trees
• Sequence models
• Example mRNA and scRNA models
• Conclusions

4
Introduction to Biosequences

• 4 nucleotides A matches T G matches C
• In RNA, U replaces T
• The NIH GenBank has 188 GB of sequence data UC
  Santa Cruz has another 128 GB
• The central dogma

http//web.mit.edu/esgbio/www/dogma/dogma.html
5
Alignments

• Alignment given two sequences, insert gaps or
  allow mismatches in input sequences to minimize a
  cost function
• Similar to edit distance
• Generalizes to n sequences
• Exploited to predict genes
• Greater similarity in protein-coding genes
• Mutated as a pair in structural RNA genes

http//hanuman.math.berkeley.edu/kbrowser
(Chakrabarti Pachter, 2004)
6
Multiple alignment

• Considering multiple sequences allows us to
  leverage the comparative genomics paradigm
• Functionally important regions of the genome are
  more likely to be conserved across species
• The converse is also true
• Genomes should be closely related
• About 5-7 species of a family (Boffelli, et. al.
  2003)
• Additional genomes increase sensitivity (true
  positives) and decrease specificity (true
  negatives)

7
Phylogenetic Trees
Durbin, et. al., 1998

• Use directed binary tree to track the
  relationships between organisms
• Each node represents the nucleotide at a
  particular position in an aligned sequence
• Current organisms are leaves of tree (observed)
• Internal nodes are the common ancestor
  (unobserved)
• Edges are speciation events and represent
  evolutionary distance as an extra parameter
• Assume each nucleotide evolves independently
  (site independent evolution)

8
Phylogenetic Tree

• Site independent model computes probability of
  independent columns
• Used for protein-coding genes
• Pairwise site dependent model computes
  probability of base-paired columns
• Used for scRNA genes

Marty Yanofsky http//www-biology.ucsd.edu/labs/ya
nofsky/images/mads/phylogenetic20tree.jpg
9
How to find a Phylogenetic Tree?

• Given n sequences, we want to find the correct
  tree topology
• Search works for small n
• Maximum likelihood choose the tree that
  maximizes the probability of the alignment

10
Biosequence analysis

• Phylogenetic trees encapsulate evolutionary time
  across sequences
• Sequence model predicts changes along the length
  of a particular sequence
• Sequence models are typically HMMs

11
Example mRNA genes

• Suppose we want to identify coding genes with an
  HMM
• Exon DNA segment that gets transcribed to mRNA
• Have states in HMM corresponding to exon regions
  (Alexandersson, et. al., 2003)
• Other types of RNA that get transcribed from DNA
  but not translated into protein are noncoding

12
Structural RNA (scRNA)

• A sequence with many self-binding sites, forming
  a stable structure.
• Implicated in regulating critical biochemical
  pathways

Michael W. King http//www.indstate.edu/thcme/mwki
ng/trna.gif
13
Example Structural RNA
Chakrabarti Ong, 2004

• Due to semi-palindromic structure, sequence model
  would be a PCFG
• Violates the site-independent assumption of
  phylogenetic trees
• Modify to allow pairwise site-dependencies in
  addition to non-matches
• Gene length can be in the thousands
• Limit the length of scRNA to constant L time
  O(L3 NL2), N length of multi-alignment

14
Example completed
Chakrabarti Ong, 2004

• Can combine HMM and the PCFG to form a supermodel
• Use a generic framework to identify mRNA, scRNA,
  and other regions

15
Phylogenetic shadowing

• Use multiple alignment of several closely related
  genomes
• Analysis of data becomes more reliable (Boffelli,
  et. al., 2003)
• More genomes reduce probability of false
  positives
• Still need closely related species to decrease
  chance of false negatives

16
Conclusions

• Phylogenetic shadowing uses a multiple alignment
  to analyze multiple genomes simultaneously,
  increasing success
• AI techniques have been proven useful in
  Computational Biology
• Still many more problems to solve

17
References

• M. Alexandersson, S. Cawley, and L. Pachter.
  SLAM Cross-Species Gene Finding and Alignment
  with a Generalized Pair Hidden Markov Model.
  Genome Research, 13 (2003) p 496--502.
  http//www.genome.org/cgi/content/abstract/13/3/49
  6
• D. Boffelli, J. McAuliffe, D. Ovcharenko, K.D.
  Lewis, I. Ovcharenko, L. Pachter, and E.M. Rubin.
  Phylogenetic shadowing of primate sequences to
  find functional regions of the human genome.
  Science, 299 (2003), p 1391-1394.
  http//www.sciencemag.org/cgi/content/short/299/56
  11/1391
• K. Chakrabarti and D.L. Ong. Computational
  Identification of Noncoding RNA Genes through
  Phylogenetic Shadowing. ACM/ISCB RECOMB 8
  (2004), poster. http//recomb04.sdsc.edu/posters/
  kushalcATuclink.berkeley.edu_168.pdf
• K. Chakrabarti and L. Pachter. Visualization of
  multiple genome annotations and alignments with
  the K-BROWSER. Genome Research 14 (2004), p
  716--720. http//www.genome.org/cgi/content/abstr
  act/14/4/716
• R. Durbin, S. Eddy, A. Krogh, and G. Mitchison.
  Biological Sequence Analysis Probabilistic
  Models of Proteins and Nucleic Acids. New York
  Cambridge University Press, 1998.