Motif Discovery in Heterogeneous Sequence Data

1
Motif Discovery in Heterogeneous Sequence Data

• Amol Prakash
• U. Washington
• Seattle

Mathieu Blanchette McGill U. Montreal
Saurabh Sinha Rockefeller U. New York
Martin Tompa U. Washington Seattle
2
Outline

• What is a motif?
• Homogeneous vs. Heterogeneous
• What makes our approach unique
• Algorithm description
• Results
• Conclusion

3
Predicting Regulatory Elements
CAGTGTTAGTCTCGACGTGAGTGGTATGAACTGGAGTTTTAGTATGATGG
TCGTACAGTGTTTCGACATGGGAAG

• Functionally important binding site for a
  protein that regulates gene expression
• Near gene
• Short Typically 6-20 nucleotides
• How can you possibly predict them?

4
Homogeneous Sequence Data I

• Input DNA sequences near co-regulated genes from
  a single organism
• Tools MEME, Consensus, Gibbs sampler,
  Projection, YMF, and many others.

CAR2 AGTCTCGACGTGAGTTTGCCTTAGGTGGTAGTTTTAAACA
GTCTCGACTAGTCTCGATCGTACAGTGTTTAGTCTTTCGACATG ARG5,
6 TTTTTTCCATTAGGTGGAGTTTTTTAGGTCTCGACAGTCTCGACTC
GTTAGTCTCGAATACAGTTTAGTCTCGAGTTTCGACATG CAR1
TCTCGACAGTTTTCACTTAGCGTTTTATCTCGAGACGTGAGTATGCCATT
AGCTGGACATG
5
Homogeneous Sequence Data II

• DNA sequences near orthologous genes
• Tools
• Multiple alignment (ClustalW, etc.), then find
  highly conserved aligned regions
• FootPrinter

CCTTGGACCAAGTCCAGCACCCTCGGGGTCGAGGAAAACAGGTAGGGTAT
AAAAAGGGCATGCAAGGACCTGCAGCCAAGCTTGCAGGTAGGGTATAA
AAAGGGCACGCAAGGGACCCCAAAAAAAGAAACTGCTCAGAGTCCTGTGG
ACAGATCACTGCTTGGCAAGAAGTGATAGATGGGGCCAGGGTATAAAA
AGGGCCCAACTCCCCGAACCACTCAGGGTCCTGTGGACAGCTCACCTAGC
TGCAAGAGGGCCCCAAAGCGCTCAGGGTCCTGTGGACAAGGGACCAGG
GTATAAAGAGGGCCCGCACAGCTGGCTCACCCCGGCTGCG
6
Heterogeneous Sequence Data

• Co-regulated genes from one species, and their
  orthologs from other species.

Rat
Mouse
Human
g1.rn
g1.mm
g1.hs
g2.rn
g2.mm
g2.hs
g3.rn
g3.mm
g3.hs
g4.rn
g4.mm
g4.hs
7
Heterogeneous Data Approach 1

• Pool everything together
• Search for statistical overrepresentation

g3.mm
g2.hs
g1.mm
g2.rn
g1.rn
g4.hs
g4.rn
g4.mm
g1.hs
g3.hs
g3.rn
g2.mm
Gelfand et al. 2000 , McGuire et al. 2000
8
Heterogeneous Data Approach 2

• Filter well conserved orthologous regions
• Search for overrepresentation in one species

g1.rn
g1.mm
g1.hs
g2.rn
g2.mm
g2.hs
g3.rn
g3.mm
g3.hs
g4.rn
g4.mm
g4.hs
Wasserman et al. 2000 , Kellis et al. 2003,
Cliften et al. 2003, Wang Stormo 2003
9
Heterogeneous Data Approach 3

• Filter overrepresentation in co-regulated
  regions.
• Search for well conserved orthologous regions

g1.mm
g1.rn
g1.hs
g2.mm
g2.rn
g2.hs
g3.mm
g3.rn
g3.hs
g4.mm
g4.rn
g4.hs
GuhaThakurta et al. 2002
10
OrthoMEME Our Approach

• An integrated approach no filtering step
• Treats orthology and co-regulation differently.
• Based on Expected-Maximization
• Does not use global alignment, which can fail on
  diverged sequences.
• Focus on two-species case

11
OrthoMEME Algorithm

• Maximization of Expected Likelihood
• Model
• As MEME, uses a profile to model the motifs in
  one genome
• Another phylogenetic profile to model motifs in
  orthologous regions.

12
OrthoMEME Profile
Profile

0.75 0.25 0 0
Rat
Human
A...
g1.rn
g1.hs
C
g2.rn
g2.hs
A
g3.rn
g3.hs
A
g4.rn
g4.hs
13
Phylogenetic Profile


Profile
A C G T
A 0.67 0.33 0 0
C 0 1 0 0
G 0 0 0
0 T 0 0
0 0

0.75 0.25 0 0
Rat
Human
A...
A...
g1.rn
g1.hs
C
C
g2.rn
g2.hs
C
A
g3.rn
g3.hs
A
A
g4.rn
g4.hs
14
Experimental Results

• Implemented and tested on various pairs of
  species
• Compared to MEME
• on single species data
• same parameters
• Results from top 3 motifs are reported.

15
Result 1 Mammals

• SRF motif
• OrthoMEME missed 2 occurrences
• MEME found none

16
Result 2 Yeast

• HAP2HAP3HAP4 motif
• OrthoMEME missed 2 occurrences
• MEME missed 4 occurrences

17
Result 3 Worm

• DAF-19 motif
• OrthoMEME missed no occurrences
• MEME missed no occurrences

18
Conclusion

• First integrated algorithm to handle
  heterogeneous sequence data.
• Focus on two species case
• Improve algorithm for multiple species.
• More experiments will help us improve the
  tool/parameters.