Information Technology Reference
In-Depth Information
discuss network inference, mostly of gene regulatory networks. Second, we will
describe how Boolean models can be used to model the dynamics of systems in
plant biology, developmental biology and immunology. In each case we will
discuss the methods followed by examples.
4.4. Boolean Inference Methods and Examples in Biology
Computational inference aims to extract causal relationships from experimental
data and to construct a network expressing these relationships. Inference of
cellular networks allows for a clearer comprehension of the inner machinery of
the cell, and when combined with modeling, can also be used to make
experimentally-verifiable predictions about cellular networks. A variety of
computational methods for network inference exists; choosing a specific
computational method depends on the nature of the data from which inferences
will be made, on the type of network under consideration, on the features of the
system one would most like to illuminate, and on the amount of computational
time available to the researcher. Boolean methods attempt to infer causative
relationships unlike empirical methods which infer associations (Fig. 4.5).
Boolean methods are usually more computationally tractable than continuous
(differential equation based) methods and are well suited to systems with limited
known information and large size.
Deterministic Boolean methods employed for the inference of gene-
regulatory networks from time-course gene expression (microarray) data seek to
define a Boolean function for each gene so that the state transition tables of
the corresponding synchronous Boolean network resemble the time-series pattern
of the system (Smolen
et al.
2000; Shmulevich
et al.
2002).
Each node's logical function is found by determining the minimum set of
nodes whose (changing) expression levels can explain the observed changes in
state of the given node in all experimental trials. Generally, an optimization
technique, such as the Coefficient of Determination (Dougherty
et al.
2000)
,
is
employed for this inference. It is possible that more than one minimum set may
be found for a particular node, and, in this case, multiple networks explain
the experimental observations. A recent analysis of the attractors of multiple
solutions found consistent dynamics over all solutions and relatively few fixed
points (Martin
et al.
2007). The search for the set of the nodes whose expression
levels explain the observed change in the expression of a particular node can be
augmented by employing probabilistic Boolean methods (Shmulevich
et al.
2002; Dougherty
et al.
2003) which incorporate uncertainty by assigning several
