Biomedical Engineering Reference
In-Depth Information
Table 4.4
Boolean regulating functions for mutated 9-Gene mammalian cell-cycle network
Gene
Regulating Function
x
1
CycD
extrac
el
lular
si
gnal
x
2
Rb
Cy
c
D
·
Cy
c
E
·
Cy
cA
·
CycB
x
3
E2F
Rb
·
CycA
·
CycB
x
4
CycE
E
2
F
·
Rb
x
5
CycA
(
E
2
F
·
Rb
·
Cdc
20
·
(
Cdh
1
·
UbcH
10))
+
(
CycA
·
Rb
·
Cdc
20
·
(
Cdh
1
·
UbcH
10))
x
6
Cdc
20
CycB
x
7
Cdh
1
(
CycA
·
CycB
)
+
Cdc
20
x
8
UbcH
10
Cdh
1
+
(
Cdh
1
·
UbcH
10
·
(
Cdc
20
+
CycA
+
CycB
))
x
9
CycB
Cdc
20
·
Cdh
1
4.3
Results
We demonstrate our predictor ranking method on two GRNs. To validate our method,
we use the mutated 9-gene mammalian cell-cycle network using synthetic gene
expression data. We use both linear and sigmoid representations for gene expression
values. From these results, we determine a predictor selection method and find that the
sigmoid representation is more accurate. Lastly, we apply both ranking and selection
method on melanoma data, assuming a sigmoid representation of gene expression
values.
4.3.1
Mutated Mammalian Cell-Cycle Network
In this experiment, we use a mutated mammalian cell-cycle network to illustrate
and validate our approach. For a normal mammal, the cell cycle is tightly con-
trolled through extracellular signals that indicate whether a cell should divide/grow
or not. These signals activate the gene CyclinD (
CycD
) which is a key gene in the
mammalian cell-cycle. Another important gene is retinoblastoma (
Rb
) which is a
tumor-suppressor when the other cyclin genes are not expressed. Another key gene
is
p
27, which when active, represses the cyclin genes, stopping the cell cycle. In the
mutated mammalian cell-cycle,
p
27 is mutated and is always off, leading to possi-
ble cell cycle activity in the absence of extracelluar signals. For the mutated 9-gene
mammalian cell-cycle network, [
16
] determined the regulating functions for genes
to be those shown in Table
4.4
. To validate our method, we will use the regulating
functions to create synthetic continuous gene expression values, on which we apply
our algorithm (linear and sigmoid) to determine predictor rankings for target genes
in the mutated network. In this setup, the actual functions and predictors are hidden
from our algorithm.
To synthesize normalized and continuous gene expression data similar to those
measured in practice, we use the following procedure. From the Boolean functions
in Table
4.4
, we create a state transition (truth) table listing all current states and next
