Reversed engineering of

1
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
A Reverter Bioinformatics Group CSIRO Livestock
Industries Queensland Bioscience
Precinct Brisbane, QLD 4067 Australia
www.csiro.au
2
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
The identification of differentially expressed
genes is a first-step non-trivial task
(RE-)START
Data generated at a rate much higher than it can
be comprehensively processed and analysed
Building a gene network is a bigger challenge
Genetic profiling is a vicious cycle
Biologically testing a network could be impossible
Understanding how the essential genes function
within a network is an even bigger challenge
Given a network, postulating a hypothesis could
be tricky (Type III Error)
Evidence of changes in network topology due to a
number of factors
3
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
Reverter et al. (2004) Joint analysis of multiple
cDNA microarray studies via multiplicative mixed
models applied to genetic evaluation of beef
cattle. J Anim Sci, 823430-3439.
Reverter et al. (2005) A rapid method for
computationally inferring transcriptome coverage
and microarray sensitivity. Bioinformatics,
2180-89.
Reverter et al. (2006) A gene co-expression
network for bovine skeletal muscle inferred from
microarray data. Phys Genomics (in press).
Reverter et al. (2005) Validation of alternative
methods of data normalization in gene
co-expression studies. Bioinformatics,
211112-1120.
The CLI Experience
Reverter et al. (2006) Simultaneous
identification of differential gene expression
and connectivity in inflammation, adipogenesis
and cancer. Bioinformatics (in press).
Reverter et al. (2005) Construction of gene
interaction and regulatory networks in bovine
skeletal muscle from expression data. Aust J Exp
Agric, 45821-829.
4
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
Reverter et al. (2004) Joint analysis of multiple
cDNA microarray studies via multiplicative mixed
models applied to genetic evaluation of beef
cattle. J Anim Sci, 823430-3439.
Reverter et al. (2005) A rapid method for
computationally inferring transcriptome coverage
and microarray sensitivity. Bioinformatics,
2180-89.
Reverter et al. (2006) A gene co-expression
network for bovine skeletal muscle inferred from
microarray data. Phys Genomics (submitted).
Reverter et al. (2005) Validation of alternative
methods of data normalization in gene
co-expression studies. Bioinformatics,
211112-1120.
Reverter et al. (2006) Simultaneous
identification of differential gene expression
and connectivity in inflammation, adipogenesis
and cancer. Bioinformatics (in press).
Reverter et al. (2005) Construction of gene
interaction and regulatory networks in bovine
skeletal muscle from expression data. Aust J Exp
Agric, 45821-829.
5
r gt 0.90
BCAR3
SGCA
GMPR2
RBBP7
B3GALT4
EEF1D
TPT1
DES
CREG
MYL2
FSTL1
RPL21
CIDEC
RPL4
RPL5
FABP4
CAV1
RPL11
ITM2B
THRSP
RPL18A
EIF3S2
NDUFC2
RPL30
GABARAP
TM4SF2
RPS17
RPS24
RPS27A
MYBPC2
GPD1
ACTA1
ENO3
PGK1
MYL1
PGM1
FN1
COL3A1
CKM
CKMT2
GPI
LDHA
CASQ1
MTATP6
PYGM
6
r gt 0.80
BCAR3
SGCA
GMPR2
RBBP7
B3GALT4
EEF1D
TPT1
DES
CSDA
CREG
MYOZ2
MYL2
FSTL1
RPL21
RPL29
CIDEC
PSMB4
FTL
RPL4
RPL5
FABP4
CAV1
RPL11
RPL17
ITM2B
THRSP
OSTF1
RPL18A
EIF3S2
CTSF
NDUFC2
SCD
RPL30
RPL31
GABARAP
TM4SF2
RARA
PPM1L
RPS17
RPS24
ACDC
RPS27A
NFE2L1
CALB3
AK1
MYBPC2
MYOM2
GPD1
ACTA1
ENO3
PGK1
SPARC
MYH1
MYL1
PGM1
FN1
COL3A1
MYL3
CKM
CKMT2
GPI
LDHA
CASQ1
ACTG2
TPM3
MTATP6
PYGM
FBP2
ATP1A2
7
STAC3
NEB
GYG
NUDT5
?
r gt 0.75
TNNC1
S100A1
BCAR3
SGCA
GMPR2
RBBP7
B3GALT4
EEF1D
ACYP2
TPT1
DES
CSDA
CREG
MYOZ2
MYL2
FSTL1
RPL21
RPL29
CIDEC
?
PSMB4
FTL
RPL4
RPL5
FHL1
FABP4
CAV1
RPL11
RPL17
ITM2B
THRSP
OSTF1
RPL18A
EIF3S2
RPL19
CTSF
NDUFC2
SCD
RPL30
RPL31
GABARAP
TM4SF2
RARA
PPM1L
?
RPS17
RPS24
ACDC
PDLIM3
RYR1
RPS27A
CSRP3
NFE2L1
CALB3
?
AK1
TTN
ACTN2
MYBPC2
MYOM2
?
GPD1
ACTA1
ENO3
PGK1
SPARC
MYH1
TT1D
FLJ31121
?
MYL1
PGM1
FN1
COL3A1
MYL3
MYH7
?
?
CKM
CKMT2
GPI
LDHA
CASQ1
SH3BGR
UQCRC1
ACTG2
?
ATP2A2
TPM3
ALDH1A1
MTATP6
PYGM
FBP2
NFKB2
PHYH
?
?
ATP1A2
TNNT3
TMOD4
NCE2
PET112L
8
STAC3
NEB
GYG
NUDT5
?
TNNC1
S100A1
BCAR3
SGCA
GMPR2
RBBP7
B3GALT4
EEF1D
ACYP2
TPT1
DES
CSDA
CREG
MYOZ2
MYL2
FSTL1
RPL21
RPL29
CIDEC
?
PSMB4
FTL
RPL4
RPL5
FHL1
FABP4
CAV1
RPL11
RPL17
ITM2B
THRSP
OSTF1
RPL18A
EIF3S2
RPL19
CTSF
NDUFC2
SCD
RPL30
RPL31
GABARAP
TM4SF2
RARA
PPM1L
?
RPS17
RPS24
ACDC
PDLIM3
RYR1
RPS27A
CSRP3
NFE2L1
CALB3
?
AK1
TTN
ACTN2
MYBPC2
MYOM2
?
GPD1
ACTA1
ENO3
PGK1
SPARC
MYH1
TT1D
FLJ31121
?
MYL1
PGM1
FN1
COL3A1
MYL3
MYH7
?
?
CKM
CKMT2
GPI
LDHA
CASQ1
SH3BGR
UQCRC1
ACTG2
?
ATP2A2
TPM3
ALDH1A1
MTATP6
PYGM
FBP2
NFKB2
PHYH
?
?
ATP1A2
TNNT3
TMOD4
NCE2
PET112L
9
Functional Annotations

• Module6
• Cytoskeleton
• Transferase activity, glycosyl
• Protein biosynthesis

• Module 4
• Muscle contraction

• Module 1
• Protein biosynthesis
• Intracellular
• Ribosome
• Structural of ribosome

• Module 5
• Structural of muscle
• Smooth endoplasmic reticulum
• Sarcomere
• Carbohydrate metabolism
• Fatty acid biosynthesis
• Energy pathways

• Module 3
• Nucleus
• Integral to plasma membrane
• Protein biosynthesis

• Module 2
• Glycolysis Creatine kinase activity
• Muscle development Tropomyosin binding
• Actin binding Myosin
• Striated muscle thick filament Magnesium ion
  binding
• Transferase activity, phosphorus

• Module 7
• Neurogenesis
• Protein biosynthesis

10
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
9 Experiments 147 Hybridizations 822 Genes 47
Conditions
11
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
Nine-Variate Mixed-Model (1,762,338 Eqs, 81
Components)
A. Reverter Sept. 2006, UAB, Barcelona, Spain
12
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
822 Genes 42,673 Significant Correlations (12.6)
Use of Partial Correlations to establish
connections in the network
X
Threshold
Z
Y
If
Connection between X and Y
and
13
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
822 Genes 42,673 Significant Correlations (12.6)
14
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
822 Genes 26 Transcription Factors
A given gene was allocated to a TF-Hub if the r
between the gene and this TF was bigger than the
r between the same gene and any other TF.
15
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
TF-Hubs Cohesiveness and Muscle Specificity
Permutation
Hyper-G
NB Entire Network, C(k) 12.6
16
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
17
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
Validation 1
Hierarchical clustering of correlation
coefficients between genes (rows) and
transcription factors (TF columns) reveals
modules that comprise clusters affected by
biologically meaningful TF.
Module 1
Module 2
Activation
Module 3
Inhibition
Module 4
18
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
Validation 2
The heat map of the r-matrix (Red positive r
Blue negative r) sorted by TF revealed
biologically sound structures.
A. Reverter Sept. 2006, UAB, Barcelona, Spain
19
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
Validation 3
A. Reverter Sept. 2006, UAB, Barcelona, Spain
20
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
Validation 3
Up-Regulated
Genes involved in food depravation have a r
structure consistent with their effect (up- or
down-regulation).
Down-Regulated
A. Reverter Sept. 2006, UAB, Barcelona, Spain
21
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
The identification of differentially expressed
genes is a first-step non-trivial task
(RE-)START
Data generated at a rate much higher than it can
be comprehensively processed and analysed
Building a gene network is a bigger challenge
Genetic profiling is a vicious cycle
Biologically testing a network could be impossible
Understanding how the essential genes function
within a network is an even bigger challenge
Given a network, postulating a hypothesis could
be tricky (Type III Error)
Evidence of changes in network topology due to a
number of factors
22
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
None
Heaps

• Complex Systems
• Expression Profiling
• Stoichiometric Analysis

Data Requirements
Expert Knowledge
Heaps
None
23
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
DIGITAL ORGANISMS

• Descendants
• Connections

• Genetic Load
• Phenotype

24
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
DIGITAL ORGANISMS
Mean Expressions
Master
Correlations
3
4
2
1
5
6
Master Genetic Load 50.0 Master
Phenotype 45.4
25
EVOLUTIONARY PROCESS
Master
26
EVOLUTIONARY PROCESS
Master
Founders
27
EVOLUTIONARY PROCESS
Master
Founders
Descendants
28
EVOLUTIONARY PROCESS
Master
Founders
Descendants
(constant population size)
Repeat at nauseum (20K Generations 1K Times)
Extreme
Extreme

• What global changes are required to generate an
  extreme phenotype?
• What are the minimal changes to generate a
  massive change in a target gene?
• What extreme changes in phenotype can be seen
  after knocking out a given gene?

29
DIGITAL ORGANISMS
Largest
Founder
Smallest
30
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications

• Simulated (Luscombe et al. 04, Nature 431308)

1
2
Armidale Animal Breeding Summer Course, UNE, Feb.
2006
31
Q Which 2 genes need to be regulated by more
than 20 to generate an extreme change to G7 and
without affecting the phenotype?
ENDOGENOUS NETWORK
Largest
G7
Smallest
32
ENDOGENOUS NETWORK
Q Which 2 genes need to be regulated by more
than 20 to generate an extreme change to G7 and
without affecting the phenotype?
Decrease G7
Increase G7
33
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications

• MYOG (Reverter et al. 05, Bioinformatics 211112
• Blais et al. 05, Genes
  Development 19, 553)

Master Genetic Load 186.9 Master
Phenotype 279.9
MYOG to be knocked out
34
MYOG NETWORK
Q What negligible and extreme changes can be
expected after knocking out MYOG?
Biological Inconsistency and/or Numerical Trickery
Biological Challenge
35
MYOG NETWORK
Q What negligible and extreme changes can be
expected after knocking out MYOG?
36
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
FINAL REMARKS

• Building a network goes beyond the identification
  of DE genes.
• Increase understanding of muscle growth.
• Grounds for new hypotheses to be postulated
  (Digital Organisms).
• Note similarities with Genetic Algorithms.
• Potentially naïve computation of Phenotype.
• to be improved with WGS studies
• Further improvements from population growth rate.
• Aggravating vs Buffering effect of connections.
• Initial results warrant further research.
• at the cutting edge of something
  incomprehensible (Peter Willadsen, CLI Chief
  Scientist).

37
Reversed engineering of gene networks for Bovine
skeletal muscle Development and Applications
ACKNOWLEDGEMENTS

• CSIRO
• Sigrid Lehnert
• Yonghong Wang
• Greg Harper
• Rachel Hawken
• Ross Tellam
• Keren Byrne
• Siok-Hwee Tan
• Nick Hudson
• Aaron Ingham
• Wes Barris
• Sean McWilliam
• Evgeny Glazov
• Abi Ratnakumnar
• Brian Dalrymple

• UNIV. ADELAIDE
• Cynthya Bottema
• Wayne Pitchford
• Adam Kister

• NSW DPI
• Paul Greenwood

• INIA SPAIN
• Clara Diaz
• Natalia Moreno

Post-Doc Opportunities!