How should we talk about biological networks or systems? Roger Brent and Jehoshua Bruck stated the problem like this:
The search for biologically relevant formalisms has a chance to greatly affect the understanding of biological function, in ways we are just starting to imagine. Today, by contrast with descriptions of the physical world, the understanding of biological systems is most often represented by natural-language stories codified in natural-language papers and textbooks. This level of understanding is adequate for many purposes (including medicine and agriculture) and is being extended by contemporary biologists with great panache. But insofar as biologists wish to attain deeper understanding (for example, to predict the quantitative behaviour of biological systems), they will need to produce biological knowledge and operate on it in ways that natural language does not allow.
Obviously, the alternative to natural language, in science anyway, is most frequently mathematical language. We need to be able to talk about biological circuits in terms of mathematics. Unfortunately, unlike physical circuits, with fairly easily quantified concepts like voltage, current, capacitance, resistance, etc., biological systems are hard to quantify. It's difficult to make all of the necessary measurements, and making the measurements in one system doesn't help you much with the next.
Biologists are looking for some way of dealing with biology that goes beyond natural language accounts, and that doesn't involve making large numbers of difficult, non-generalizable measurements.
One way of thinking about systems is to ask why a particular regulatory network is rigged up the way it is. We have to be careful asking this kind of question, because one answer is simply that the system is the result of its contingent evolutionary history. But it can be fruitful to ask what the functional advantages are of rigging up feedback one way rather than another.
One scientists who has been thinking about this issue for decades is Michael Savageau. Here, I turn to an older paper of his
: "Comparing systemic properties of ensembles of biological networks by graphical and statistica methods," published with Rui Alves in 2000 in the journal Bioinformatics
Alves and Savageau start with some straightforward criteria for comparing biological systems:
The only rigorous way to characterize and compre alternative biological designs for a particular class of systems is through the use of mathematical models and quantitative methods of analysis. In pursuing these goals, we must address three crucial issues. First, biologically meaningful behaviors must be identified (or, as is more commonly the case, hypothesized) and characterized by quantitative measures. Second, a representation of the alternatives must be capable of describing the phenomena of interest in quantitative terms. Third, comparisons will require analyses that explore a range of parameter values and use statistical methods to evaluate the results.
One of the critical points here is the fact that, as we analyze a gene regulatory network or signal transduction pathways, we must choose of hypothesize meaningful biological behaviors. There is no automated way of discovering which aspect of a system is biologically important. Alves and Savageau put it this way:
There is no prescription for discovering those biological behaviors that are based on natural selection of those that occur at random with high probability. The behaviors that are important characteristics of a given biological system can only be discovered by experimental means. Hypotheses must be generated and tested in each case, and this process will vary considerably according to the systems being studied. The behavioral repertoire of nonlinear systems can be quite diverse, including saturation, thresholds, memory, time delays, synchrony stable limit cycles, and strange attractors.
I'm highlighting this point because this is something that often gets lost in genomics-driven systems biology. Inferring network connections from genomic data may be great for putting together the wiring diagrams, but without directed experimentation on a given system, we can't say much about how a system is carrying out its function.
The biggest implication of this idea that there is no automatic or prescribed way of finding important functional features of systems is that our ability to generalize in systems biology may be limited. It's not always true that, as a famous molecular biologist once put it, that what's true for E. coli
is true for an elephant.
In terms of molecular biology, model organism research has been amazingly successful, but I'm worried that, when it comes to systems biology, we're forgetting the limits of model organisms. Much of what's true about yeast cell division is true about mammalian cell division, but at some point, these systems begin to differ significantly. For molecular biologists, that point is relatively easy to find: you look for the homologous genes. But for systems biologists, it's more difficult to know where you hit the point of diminishing returns. Does the transcriptional regulatory circuit that controls the induction of early G1-phase genes in yeast resemble the mammalian transcriptional circuit? I have no idea.
In some spectacular cases, similar regulatory circuits composed of non-homologous genes are found in very different organisms, but as far as I can tell, this kind of thing is the exception. What this means is that systems biologists, especially those who work on experimentally tractable model organisms (which are great, because we can make the necessary measurements), need to be careful as they frame their problems, and not become excessively focused on the local specifics of their particular model system.
Read the feed: