Biology Reference
In-Depth Information
Reverse engineering is a critical step in the systems biology paradigm that has
pervaded the biological sciences in recent history. The modeling process starts with
a biological system under study. Given a question or hypothesis about the system,
the researcher designs experiments to probe the system in hopes of addressing the
question or hypothesis. A model is then built from the data resulting from the exper-
iments. Predictions or simulations are extracted from the model and then compared
against the original system. The step “from data to a model” is one that uses reverse
engineering. This paradigm differs from the previous “reductionist” paradigm that
dominated twentieth century biology: the parts of a system were of importance for
study in reductionism, whereas the system itself is of interest in systems biology.
Generally speaking, reverse engineering in systems biology aims to recover the
network topology and dynamics of a network from observations. A model is built to
fit the given observations and the topology and/or the dynamics of the network can
be inferred from the model. Network topology refers to the physical structure of the
network, that is, how the components in the network are connected. It is often encoded
as a directed graph, or a wiring diagram, where vertices represent the components of
the network (genes, proteins, signaling molecules, etc.) and a directed edge is drawn
between two vertices if there is an interaction between the associated components
(regulation, synthesis, activation, etc.). Dynamics refer to the behavior of the network
over time, that is the time evolution of the network processes. The dynamics of a
network is also depicted as a graph; in the case of finite dynamical systems, which
are considered here, the graph is finite and directed. The so-called state space graph
is comprised of vertices representing network states ( n -tuples) and a directed edge is
drawn between two vertices A and B if the network advances from state A to state B.
See Section 3.2 for definitions.
Gene regulatory networks are often modeled using collections of mathematical
functions in which each molecular component is assigned a function that formalizes
the dynamics of the component [ 5 , 6 ]. From these functions the wiring diagram and
state space can be constructed and the structure and behavior of the GRN can be
analyzed. A challenge for molecular geneticists is to identify causal links in the
network. Identification and control of these links are important first steps in repairing
defects in regulation. While there is a growing amount of data being collected from
such networks, control of GRNs requires knowledge of the topology and dynamics
of the GRN.
Mathematical methods to reverse engineer GRNs are diverse and draw from statis-
tics, graph theory, network theory, computational algebra, and dynamical systems
[ 7 , 8 ]. The performance of these methods intrinsically depends on the amount and
quality of data provided [ 9 ]. In practice, there is insufficient data to uniquely infer
a model for a GRN and the number of models that fit the data may be considerably
large. An area of continual growth is the development of methods to select biologi-
cally feasible or likely models from a pool of candidate models. There are numerous
strategies for model selection. For example, some methods restrict the space of likely
wiring diagrams to those that have few inputs per vertex or whose in-degree distribu-
tion follows a power law, features which are consistent with what is believed about
 
Search WWH ::




Custom Search