Promoter Discovery: A Correlation Mining Approach

1
Promoter Discovery A Correlation Mining Approach

• Yi Lu
• Department of Computer Science
• Wayne State University

2
Outline

• Introduction
• Related Work
• Problem Definition
• Correlation Mining
• Conclusion and Future work

3
Introduction

• Central Dogma
• Gene Expression

4
Introduction

• The promoter region (a set of transcription
  binding sites) of the gene acts as light switch.
  It signals when to turn the gene on and off.
• We are interested in the relationship between the
  promoter region and gene expression. i.e. what
  kind of binding sites determine whether a gene is
  expressed or not?

5
Introduction - Microarray
Microarray chips
Images scanned by laser
Gene Value D26528_at
193 D26561_at 70 D26579_at
318 D26598_at 1764 D26599_at
1537 D26600_at 1204 D28114_at
707 H29189_at 899 G29183_at 9210
Datasets
D1 D2 D3 D4..
D26528_at D26561_at D26579_at D26598_at D26599_at
D26600_at D28114_at .. ..
6
Introduction

• Transcription factor binding sites (motif) in
  promoter region should explain changes in
  transcription.

Time Course genes
AGCTAGCTGATTGTGCACACTGATCGAGCCCCACCATAGCTTCGTTGTG
CGCTATATATTGTGCAGCTAGTAGAGCTCTGCTAGAGCTCTATTTGTG
CCGATTGCGGGGCGTCTGAGCTCTTTGCTCTTTTGTGCCGCTTTTGAT
ATTATCTCTCTGCTCGTTTGTGCTTTATTGTGGGGGTTGTGCTGATTAT
GCTGCTCATAGGAGATTGTGCGAGAGTCGTCGTAGTTGTGCGTCGTCG
TGATGATGCTGCTGATCGATCGTTGTGCCTAGCTAGTAGATCGATGTT
TGTGCAGAAGAGAGAGGGTTTTTTCGCGCCGCCCCGCGCTTGTGCTCG
AGAGGAAGTATATATTTGTGCGCGCGCCGCGCGCACGTTGTGCAGCTGA
TGCATGCATGCTAGTATTGTGCCTAGTCAGCTGCGATCGACTCGTAGC
ATGCATCTTGTGCAGTCGATCGATGCTAGTTATTGTTGTGCGTAGTAG
TGCTTGTGCTCGTAGCTGTAG
AGCTAGCTGATTGTGCACACTGATCGAGCCCCACCATAGCTTCGTTGTG
CGCTATATATTGTGCAGCTAGTAGAGCTCTGCTAGAGCTCTATTTGTG
CCGATTGCGGGGCGTCTGAGCTCTTTGCTCTTTTGTGCCGCTTTTGAT
ATTATCTCTCTGCTCGTTTGTGCTTTATTGTGGGGGTTGTGCTGATTAT
GCTGCTCATAGGAGATTGTGCGAGAGTCGTCGTAGTTGTGCGTCGTCG
TGATGATGCTGCTGATCGATCGTTGTGCCTAGCTAGTAGATCGATGTT
TGTGCAGAAGAGAGAGGGTTTTTTCGCGCCGCCCCGCGCTTGTGCTCG
AGAGGAAGTATATATTTGTGCGCGCGCCGCGCGCACGTTGTGCAGCTGA
TGCATGCATGCTAGTATTGTGCCTAGTCAGCTGCGATCGACTCGTAGC
ATGCATCTTGTGCAGTCGATCGATGCTAGTTATTGTTGTGCGTAGTAG
TGCTTGTGCTCGTAGCTGTAG
AGCTAGCTGATTGTGCACACTGATCGAGCCCCACCATAGCTTCGTTGTG
CGCTATATATTGTGCAGCTAGTAGAGCTCTGCTAGAGCTCTATTTGTG
CCGATTGCGGGGCGTCTGAGCTCTTTGCTCTTTTGTGCCGCTTTTGAT
ATTATCTCTCTGCTCGTTTGTGCTTTATTGTGGGGGTTGTGCTGATTAT
GCTGCTCATAGGAGATTGTGCGAGAGTCGTCGTAGTTGTGCGTCGTCG
TGATGATGCTGCTGATCGATCGTTGTGCCTAGCTAGTAGATCGATGTT
TGTGCAGAAGAGAGAGGGTTTTTTCGCGCCGCCCCGCGCTTGTGCTCG
AGAGGAAGTATATATTTGTGCGCGCGCCGCGCGCACGTTGTGCAGCTGA
TGCATGCATGCTAGTATTGTGCCTAGTCAGCTGCGATCGACTCGTAGC
ATGCATCTTGTGCAGTCGATCGATGCTAGTTATTGTTGTGCGTAGTAG
TGCTTGTGCTCGTAGCTGTAG
R(t2)
7
Related work

• Cluster gene expression profiles
• Search for motifs in promoter regions of
  clustered genes

8
Related work

• Clustering
• partition the N genes to a set of disjoint groups
  so that the expression profile of genes in same
  group have high similarity to each other and the
  expression profile of genes in different groups
  are dissimilar to each other.
• Most widely used algorithms K-means clustering,
  hierarchy clustering algorithms.
• Genetic K-means algorithms (Lu et al. 2003,
  2004).

9
Related work

• Motif discovery after clustering
• given a set of upstream sequence of genes which
  are co-expressed, find subsequences that are
  overrepresented and are significant to be
  separated from other subsequences
• MEME, Gibbs Sampling, Winnower algorithms.
• PDC algorithm (Lu et al. 2006)
• Usually have high false positive rate

10
Motivation

• Researches have indicated that multiple
  transcription factor binding sites are involved
  into each transcription process. This lead us to
  study the Modules (a pair of motifs) instead of
  Motifs.

11
Motivation

• Not all genes contain the same motif cause the
  same gene expression change.
• Not all genes with same gene expression change
  contains same motif.

12
Problem Definition

• Given a list of genes, and corresponding module
  present information, gene expression information,
  find the relationship between module and gene
  expression, i.e. which modules or module
  combinations may relate to the gene expression
  change.
• M1 M2 gt increase gene expression change from
  Day 1 to Day 4

13
Method - Quantify Gene Expression
14
Method - Quantify Gene Expression
15
Method Generate Frequent Module Set

• Frequent module sets (occurrence gt2)
• M1(4), M2 (3), M3 (2)

, M4(1)
M1M2 (3), M1M3 (2)
, M2M3 (1)
M1M2M3(1)
16
Method Generate Frequent Gene Expression Set

• Frequent gene expression sets (occurrence gt2)
• E1 (2), E1- (0), E2 (1), E2-(3), E3 (0),
  E3-,(2),
• E1E2-(1), E1E3-(1), E2-E3- (2)

17
Correlation Measure Contingency Table

• The relation between u and v in the pair (u,v)

18
Liddell Measure

• Liddell ( 21-10)/(22) 0.5

19
Method Correlate Module Set with Gene
Expression Set

• Minimize module set
• Maximize gene expression set
• Minimum Liddell value is set to 0.5/-0.5, then
  the result sets
• M2 -gtE1
• M2 -gt (E2- E3-)
• M3 -gtE2- E3-

20
Result on Spermatogenesis

• Spermatogenesis is the biological process related
  to formation of sperm. Two gene expression data
  sets are downloaded from GEO (Gene Expression
  Omnibus).
• The time course of one dataset ranges from day 0,
  3, 6, 8, 10, 14, 18, 20, 30, 35, and 56. And the
  other ranges from day 1, 4, 8, 11, 14, 18, 21,
  26, 29, and 60.

21
System Workflow

• GEO Gene Expression Omnibus
• DBTSS DataBase of Transcriptional Start Sites
• TRANSFAC the Transcription Factor database
• JASPAR The high-quality transcription factor
  binding profile database

22
Conclusion

• Not only same module combination result, but also
  the same genes that contain the module
  combinations have been pulled out between the two
  datasets.
• The promoter detected using our approach
  statistically shows significance than random
  generated datasets.
• Some promoters found by our approach are
  confirmed by literatures.

23
Future work

• The concordance between the two gene expression
  datasets downloaded from GEO are low, new method
  to reconcile the difference between two data sets
  is needed.
• Motifs found by different algorithms are
  overwhelming, we may incorporate the weight
  matrix and gene ontology to identify the
  significant ones.

24
References

• Gene Expression Clustering
• Yi Lu, Shiyong Lu, Farshad Fotouhi, Youping Deng
  and Susan Brown, "FGKA A Fast Genetic K-means
  Clustering Algorithm", in Proceedings of the 19th
  ACM Symposium on Applied Computing, Nicosia,
  Cyprus, March, 2004.
• Yi Lu, Shiyong Lu, Farshad Fotouhi, Youping Deng,
  and Susan Brown, Incremental Genetic K-means
  Algorithm and its Application in Gene Expression
  Data Analysis, International Journal of BMC
  Bioinformatics, 5(172), October, 2004.
• Motif Discovery
• Yi Lu, Shiyong Lu, Farshad Fotouhi, Yan Sun and
  Zijiang Yang, PDC Pattern Discovery with
  Confidence in DNA Sequences, In the proceedings
  of the IASTED International Conference on
  Advances in Computer Science and Technology (ACST
  2006), Puerto Vallarta, Mexico, January, 2006
• Motif Extraction, Module Integration
• Adrian E. Platts, Yi Lu, Stephen A. Krawetz,
  K-SPMM, an Online System for Data Mining
  Regulatory Elements from Murine Spermatogenic
  Promoter Sequences, presented in 2006 Great
  Lakes Mammalian Development Meeting, Toronto,
  March 3-5 2006.
• Yi Lu, Adrian E. Platts, Charles G. Ostermeier,
  Stephen A. Krawetz, A Database of Murine
  Spermatogenic Promoters Modules Motifs,
  Submitted to Journal of BMC Bioinformatics for
  publication.
• Correlation Mining
• Yi Lu, Adrian Platts, Shiyong Lu, Jeffrey L. Ram
  and Stephen Krawetz, "Correlation Mining to
  Reveal the Regulation of Transcription Factor
  Binding Site Modules", 4th Great Lake
  Bioinformatics Retreat, Frankenmuth, Michigan,
  August, 2005.
• Yi Lu, Adrian Platts, Shiyong Lu, Jeffrey L. Ram
  and Stephen Krawetz, Mining of Correlation
  Between Transcription Binding Sites and Gene
  Expression Profiles, In preparation.

25
(No Transcript)
26
Acknowledgements

• Dr. Shiyong Lu
• Dr. Stephen Krawetz
• Mr. Adrian Platts
• Dr. Jeffrey Ram
• Dr. Youping Deng

27
Questions?