Biology Reference
In-Depth Information
FIGURE 22.4
Data integration and model construction
strategies. Flow charts show possible ways of data integra-
tion, construction of SIN and BRN-based models and their
relationships. While the construction of SINs, such as Boolean
networks and models, requires data integration, construction
of BRNs requires preliminary analysis of the networks. BRNs
display network modularity; simplified to the level of nodes
with known input/output characteristics, BRNs may be
consequently 'integrated back'
into larger networks and
models.
¼
interest. Other miscellaneous databases include StemDB
(http: //
www.stemdb.org
), which holds stem cell-related
information
functions F
, f
n
}. A Boolean function f
k
is a logical
function operating on the values of upstream nodes regu-
lating the activity of node k via Boolean operators such as
'AND', 'OR' and 'NOT'. At each time step the nodes can
be updated in a synchronous or asynchronous manner.
Synchronous updating simply assumes that the reactions in
the network G have similar timescales. Therefore, a node is
updated after all rules have been applied and all nodes are
updated simultaneously; alternatively, asynchronous
updating takes temporal ordering into account and can be
categorized as either undeterministic, stochastic asynchro-
nous, such as randoming order updates
[85]
, or determin-
istic updating
[86]
. The total number of states of a network
is finite (2
N
for a network G consisting of N nodes). For
synchronous and deterministic asynchronous updating, the
system typically arrives at steady states relatively quickly
characterized by either a limited cycle or a fixed-point
attractor. In contrast, states outside the attractors are tran-
sient and unstable. For each attractor, the set of all transient
states leading to that attractor constitute the basin of
attraction similar to the behavior of continuous variables
dynamical systems (see
Figure 22.5
C,D).
Boolean network modeling can capture the collective
behavior of sophisticated regulatory networks and has been
applied to explore several complex biological systems. A
Boolean network has been developed to simulate the yeast
cell cycle and to predict cell cycle events
[87]
. A probabilistic
Boolean network has been successfully applied to analyze the
dynamical behavior of a subnetwork consisting of 15 genes in
human glioma
[88]
. Based on the analysis of joint steady-state
probabilities for Tie-2, NF
{f
1
,
.
(mRNA expression
profiles,
antibodies,
primers and protocols).
New provisional gene networks for mouse ESCs are
constantly being generated and updated
[14,15,20,34,36,72]
.
However, these networks are often focused on various
specific aspects of mESC biology, constructed from data
obtained under different conditions (e.g., different knock-
downs or differentiation conditions) and carry substantial
disagreements. Similarly designed genome-wide studies of
the core pluripotency factors protein
protein interactions
(interactomes)
[80,81]
may produce only 20
e
40% of true
positives that are consistent between the studies
[82]
.
Comparisons of large numbers of genome-wide datasets
may produce even weaker agreements.
On account of the enormous amount, diversity and
ambiguity of currently available data, one prominent goal is
to integrate various kinds of genome-wide data across
multiple regulatory layers in order to build more consistent
GRNs and predictive quantitative models.
e
Simple Binary Models for Complex
Gene Networks
Abstract information regarding regulatory interactions
between genes can be approximated by rather simple
Boolean on/off switches; Boolean networks are among the
common computational approaches to model the dynamics
of the GRNs
[28,31,83,84]
. A Boolean network can be
represented by a directed graph G(V, F), as in
Figure 22.5
A.
G is defined as a set of vertices (nodes) V
B, the model
predicted function for Tie-2, a receptor tyrosine kinase
involved in tumor development. Frequently, Boolean
networks are used to infer the underlying structure of GRNs
from high-throughput
k
BandTGF-
b
3, NF
k
¼
{x
1
,
, x
n
}
.
connected by a set of Boolean functions F
, f
n
}.
Under the Boolean framework, each variable can take
binary values of 1 representing the 'ON' or active state, or
0 representing the 'OFF' or inactive state. The state of each
node at time t is determined by the value of other nodes in
the previous time step according to a list of Boolean
¼
{f
1
,
.
time series of microarray data
[89
92]
. The REVEAL algorithm developed by Liang was
among the first formalism to infer Boolean model structure.
The algorithm uses mutual information to determine the
e
