Phylogenetic Models for Motif Detection

1
Phylogenetic Models for Motif Detection

• Pradipta Ray
• ( joint work advised by Eric Xing )

2
What

• Repeating, roughly conserved genetic sequence of
  biological sequence
• Pure pattern matching techniques give you a
  large number of false matches low precision
• Best modelled as semi supervised

3
The Problem Definition

• Motif formalism
• Positional Weight Matrix
• Sequence of 4-nomials
• Supervised Simple MLE for multinomial

4
Traditional Approaches

• Unsupervised
• Gibbs Sampling
• Expectation Maximization
• De novo detection

5
Knowledge is power

• Motifs are conserved sites
• Given aligned sequences, we may choose suitable
  regions

6
A look at motif evolution

• Work by Krietman et al
• Functional evolution granularity of the
  evolving unit is larger than the nucleotide

7
Phylogenetic Trees
( T , L , P , M ) T ( V, E ) Tree topology L
E ? R Edge lengths P Multinomial parameters
for initial draw M CTMM parameters
8
Phylogenetic HMM
A G G A A
C1
P1
9
The mixture of trees model
10
A G G A A
11
(No Transcript)
12
Learning

• Learning the evolutionary trees
• Learning the annotation trees

13
Inferencing the Tree
?
?
?
?
14
Usage

• Framework for answering questions about evolution
  of macro-entities
• A specific and highly significant case would be
  that of motif finding

15
Conclusion

• Currently being implemented
• To be tested with data from the Drosophilae
  species