A Glossary for Systems Biology
Contents
Index - Glossary -
Modularity
A Closer
System
Submit - Comment
Model
The word stems from the Latin 'modus', meaning 'measure'
or 'type'.
Historically, the term first meant scale models of the original, i.e.
a model of something. It is still used that way
in terms like 'model airplane'.
Later it designated analogies that were used to explain complex concepts
by comparison with well-known objects, like the atomic model
explained by analogy to a planetary system. These kinds of models
enable us to postulate, extrapolate and propose hypotheses which can
then be tested against reality. In these cases the term is used in
the meaning of a model for something [27].
The kind of model we talk about in systems biology is still
a model for something, but at an even more abstract
level.
Classical biologists use model as a short form of 'model
organism'. By that they mean an organism that is sufficiently similar
in all respects important for a certain experiment that it can be
used instead of the organism that the results of that experiment are
ultimately meant for. A typical example is the use of laboratory animals
for experiments, for example guinea pigs (real or metaphorical), mice,
pigs, flies, worms, yeast, etc., to make predictions about human diseases
and their possible treatment.
Molecular biologists use the term model in a way similar
to that of systems theorists; they understand it as a formal description
of connections and dependencies between distinct parts of a metabolic
pathway. The difference is that they usually use verbal descriptions,
not mathematical equations and that their models usually
are limited to qualitative descriptions of connections, without quantitative
influences.
For systems theorists, model means a system of equations,
with initial and boundary conditions, that (ideally) describe the
qualitative and quantitative behavior of a system, including its dynamics,
over the full range of values.
The fundamental difference between mathematical models of
molecular biologists and systems theorists lies in the range of dynamical
behavior that they expect the model
to be able to represent. Most models of biological systems
still only cover steady-state properties of a
system (static), or are valid for a linearization of the system
around a certain steady state (local
dynamic).
This is due to the nature of measurements taken in biology: samples
have to be taken and analyzed for the compounds of interest. This
sampling process has a limited resolution in time - the maximum frequency
at which samples can be taken and analyzed. This time constant often
does not allow to get series of measurements that represent the (sometimes
quite fast) dynamic behavior of intra-cellular processes, but rather
are limited to states a system stays in for a sufficiently long (i.e.
measurable) time.
Biological systems are typically non-linear, resulting in complex
dynamic behaviors.
The stable states of a system are those easiest to identify by simple
observation. They are also the states best suited to take measurements,
since values do not change quickly, accommodating traditional (slow)
measurement techniques. Transitions between stable states are often
quite fast, defying measurement with traditional methods. Consequently,
models were built which consider transitions to be simple switching
between stable states. They are often called 'connectionist'
models because they only consider the fact that a connection exists
between two states, but not the dynamics of the transition. These
dynamics contain additional information about a system, though, so
there is an interest to obtain it and incorporate it into models.
A first step can be to consider local dynamics, i.e. the
dynamics within a small area around a steady state. Local dynamics
can often be extracted from measurement data, giving a first indication
of transition dynamics. Such information could be the direction the
states begin to change in, or whether the changes start slowly and
accelerate or are fast from the beginning.
Such knowledge can be represented in linear models, which are valid
around that specific steady state. Such models would be a linear approximation
of the real non-linear dynamics of the original system.
Recent improvements to traditional measurement techniques and newly
developed technologies enable measurement of the fast dynamic changes
during transitions. They provide a wealth of empirical data which
can be used to improve existing models. Those improved models will
usually be non-linear, making analysis more difficult. They will be
able to represent a systems global dynamics, though.
Another problem that is closely connected to the dependence on empirical
data described above is the specificity of the resulting models. Due
to their empirical nature they are in most cases limited to represent
a special cell line's behavior. Some might show qualitative connections
and dependencies of proteins, but cannot explain the quantitative
behavior of these connections. Others might show the qualitative behavior
of a metabolic pathway for a selected cell line, might even allow
ball-park estimations of quantitative dependencies, but are unable
to cross the border to even closely related cell types. This is due
to the fact that different cell lines might act identically in a qualitative
way but in a completely different quantitative dimension; or even
change their qualitative behavior completely along with concentration
dimensions while working along the exact same interconnection pathways
of proteins (modularity) [58].
As already mentioned above, advances in measurement techniques and
connected fields bring the hope that empirical data might soon be
available that will allow more complex models to be developed.
But this is not necessarily a good development in all respects.
These new models will be very detailed, as they will be based
on a wealth of data, and that might just as well hinder effective
research in some areas. Depending on the goals of that research, ``it
is important to carefully consider the purpose of model building:
Whether it is to obtain an in-depth understanding of system behavior
or to predict complex behaviors in response to complex stimuli, we
must first define the scope and abstraction level of the model''
[34].
If, for example, the aim of a series of simulations is to investigate
the long-term changes that a certain stimulus induces (e.g. in a case
of adaptation), then the short-term fluctuations
that occur in between the stimulus and the adaptation might
be of little or no interest for the research aim [54].
Nevertheless, they will still have to be simulated and will take up
a lot of computing power and time, even though they could be irrelevant.
A model with the right abstraction level would leave out
the short term fluctuations and only show effects in the relevant
time scale.
The problem with this idea is that it is not all that easy to decide
which effects can safely be left out of a simulation for a specific
purpose without compromising the integrity of the simulation result.
Functional dependencies of the different parts of a pathway are the
critical issue here [54]. A modular concept (modularity),
based on realistic functional modules in the biological pathways,
could be an important tool in classifying which elements are important
for a certain research goal and which are not.
``The problems of applying systems theory in biology'' include
``the difficulty of building precise and yet general models''
[62]. Therefore, it will be equally important to
make progress not only in obtaining base data for developing new and
better models but also in ``studies on classifications
and comparison of circuits'' within biological systems,
to get a feel for the ``richness of design patterns used'' and
their evolutionary connections. The hope behind this kind of research
would be to find ``a possible evolutionary family of circuits''
or ``a periodic table for functional regulatory circuits''
[34].
Such a 'family of circuits' might form the basis for a family of modules
and modular models that will allow selective simplification
of models (or rather adaptation
of a model's level of abstraction) to fit their level of abstraction
to the purpose of the research they are used for (modularity).
Next: Modularity
Up: A Closer
Previous: System
  Contents
  Index
Glossary Submit Comment
|