Cis-regultory module

1
Cis-regultory module

• 10/24/07

2
TFs often work synergistically
(Harbison 2004)
3
Combinatorial control
4
l-phase
E coli
lytic growth
lysogenic growth
(source Gary Kaiser)
5
l-operon
cro
cI
OR
6
l-operon
lysogenic growth
on
off
cro
cI
OR
7
l-operon
lytic growth
off
on
cro
cI
OR
OR1
OR2
OR3
8
l-operon
lysogenic
Pol II
cro
cI
9
Cis-regulatory module (CRM)

• A CRM is a DNA segment, typically a few hundred
  base pairs in length containing multiple binding
  sites, that recruits several cooperating factors
  to a particular genomic location
• Ji and Wong (2006)

10
Statistical Methods

• Predict modules when the motifs are known.
  (simpler)
• LRA, by Wasserman and Fickett (1998)
• Predict modules when the motifs also need to be
  discovered. (more difficult)
• CisModule, by Zhou and Wong (2004)
• EMCModule, by Gupta and Liu (2005)

11
LRA
12
LRA
Basic idea True regulatory regions are likely to
have multiple motif sites.
Probability for being regulatory
13
LRA
Probability for being a regulatory region
regression coefficient
highest motif matching score within a given
sequence

• Training data contain a subset of known
  regulatory and control regions.

14
Application skeletal-muscle gene regulation

• 5 muscle-specific TFs are known
• Mef-2, Myf, SRF, Tef, Sp-1
• 29 regulatory regions are known.
• Can we predict the regulatory regions just from
  sequence motif information?

15
Computational Procedure

• Motif matrices are identified by Gibbs sampling
  using sequence information from the 29 regulatory
  regions.
• For some TF, motifs cannot be found by the de
  novo approach. Use literature motifs instead.
• Top two matching scores for each TF are included
  as covariates.
• Apply LRA model. Use leave-one-out
  cross-validation to evaluate model performance.

16
Results

• Single motifs are highly non-specific.
• Simple multi-sites analysis improves specificity
  at the cost of reducing sensitivity.

17
Results

• Single motifs are highly non-specific.
• Simple multi-sites analysis improves specificity
  at the cost of reducing sensitivity.

18
Results

• Single motifs are highly non-specific.
• Simple multi-sites analysis improves specificity
  at the cost of reducing sensitivity.
• Logistic regression further improves specificity
  at reduced cost for sensitivity.

19
Limitations of LRA

• Motifs must be known in advance.
• When known regulatory sequences are few, it is
  difficult to identify motifs by using traditional
  methods.
• Objective
• Integrating motif discovery and module finding in
  a single statistical model.

20
De novo module identification

• Two tasks
• Identify TF motifs
• Identify CRMs.

21
Why module approach can help motif discovery

• Due to poor specificity, a short sequence can be
  enriched simply by chance.
• The probability for random matches is much
  smaller for motif co-occurrence.

22
cisModule

• Basic idea
• A two-level hierarchical mixture model (HMx).
• Level 1 modules ? sequences

(Zhou and Wong 2004)
23
cisModule

• Basic idea
• A two-level hierarchical mixture model (HMx).
• Level 1 modules ? sequences
• Level 2 motifs ? modules

(Zhou and Wong 2004)
24
HMx Model as a Stochastic Process

• Treat HMx model as a stochastic machinery to
  generate sequences.
• From the first sequence position, make a series
  of random decisions of whether to initiate a
  module of length l or generate a letter from the
  background model.
• Inside a module, If a site for the kth motif was
  initiated at position n, then generate wk letters
  from its PWM and place them at n, nwk-1,
  otherwise generate a letter from the background.
• After reaching the end of the current module,
  decide whether sampling from the background or
  initiating a new module.

(Zhou and Wong 2004)
25
Model inference Gibbs sampling
given model parameters, update module/motif
locations
26
An numerical experiment

• Merge the 29 regulatory regions with a set of
  sequences randomly selected from ENSEMBL
  promoters.
• Test the ability of cisModule to identify motifs
  under noisy environment.

27
Results
28
Limitations of CisModule

• The length of module, and number of motifs are
  externally provided.
• Convergence time could be slow. Multiple cycles
  are needed each starting from a new seed.
• Assuming that combinations of different motifs
  are independent.

29
EMCModule

• Gupta and Liu (2005) developed a similar approach
  called EMCModule.
• Main difference
• They use the collection of literature motifs as
  initial seeds for motif discovery.
• Their method improves the convergence speed.
• Their definition of CRMs are a little different
  the number of motifs are fixed within one module,
  but the order of and distance between different
  motifs can be varied.

30
Further issues

• Comparative genomic approach can also be
  incorporated into module discovery. (Zhou and
  Wong 2007).
• The modules identified by these methods can be
  viewed as belonging to one type. New methods
  need to developed to discover multiple module
  types.
• While module-based approach is helpful for
  finding cooperative motifs, it may hurt discovery
  of single motifs.

31
(Yuh et al. 1998)
32
(Yuh et al. 1998)
33
(Yuh et al. 1998)
34
(Yuh et al. 1998)
35
Reading List

• Wasserman and Fickett (1988)
• LRA. One of the first work on cis-regulatory
  modules.
• Zhou and Wong (2004)
• cisModule. A statistical method to identify cis-
  regulatory modules without knowledge of motif
  information.
• Yuh et al. (1998)
• An influential biological paper on how
  information can be integrated from different
  modules to regulate gene expression.