Over at Ars Technica’s Nobel Intent, Robert Fortner takes a long, insightful look at synthetic and systems biology:
So far, neither systems nor synthetic biology has been able to find an unambiguous anchor point from which to spin a web of equations enclosing living matter. Instead, systems biology models grow and grow, but the end result is a molecular anatomy of overwhelming detail. Without models, making use of this mass of knowledge for medical applications becomes very difficult. Although synthetic biology begins at the bottom and looks up, it eventually comes face to face with the same complexity that systems biology sees peering from the top down. In either direction, the more we look, the more we find. In the pincer movement of systems and synthetic biology on complexity, complexity is winning.
Complexity certainly kicked my ass this summer.
After nearly two years of work, I set aside a project without published results, because the ostensibly simple system I was trying to model turned out to be more complex than I hoped. Fortner writes about similar difficulties encountered by systems and synthetic biologists, and he rubs salt in these wounds by pointedly quoting from hype-filled papers published 5-7 years ago.
The basic perspective Fortner lays out is this: Systems biologists hoped that masses of omic data would enable them to build predictive models of entire cells, or at least of large networks within those cells. Most of these modeling efforts were focused on yeast, a relatively simple, single-celled eukaryote, with the rationale being the fact that we have a ton of yeast data, and if we can’t model yeast, then we have no hope of modeling human cells. Ten years later, progress has been minimal, and Fortner quotes a number of pessimistic sentiments expressed by some big names in the field.
If systems biologists couldn’t get a grasp on complex biological systems by trying to put terabytes of data into big models, perhaps the route to success lies with the bottom-up approach of synthetic biology, which focuses on building small biological systems from scratch. However, synthetic biology has also hit a wall, and the biological gadgets being engineered are no more complex than what people were making back in 2001.
Much of the problem is perhaps that systems and synthetic biologists have no theories on which to base their models and engineering designs:
Engineering under such circumstances presents serious obstacles. Because there is no mathematical model for yeast, for example, “It’s like trying to design an airplane with a very limited understanding of fluid dynamics,” as Jean Peccoud, a synthetic biologist at Virginia Tech, put it. Aeronautical engineers have the Navier-Stokes equation; biological engineers have experimentation, trial-and-error. Continued Peccoud: “the cell is the environment and the difficulty comes from the fact that we don’t understand the physics of this environment.”
To me this suggests that we should be pursuing an angle that wasn’t covered in Fortner’s piece. Instead of trying to reverse engineer huge natural systems, or build small artificial ones completely from scratch, we need to focus on learning from small, natural systems and get the best of both worlds.
The fact is, we’re more likely to learn something useful by building a comprehensible model of a small system than we are from a gigantic model that can spit out predictions but not really give us any theoretical insight.
Systems biologists love to draw analogies between biology and physics and engineering, but we seem to miss the lessons of those fields. The reason the Navier-Stokes equation is so damn cool is because it’s a relatively simple equation that can produce complex dynamics. The way to tackle complexity is not to study complex models. We understand complexity by focusing on the simple, underlying interactions that produce it.
What that means in practice is that we should study simple biological systems that have interesting behaviors. Howard Berg has been doing this successfully for decades in his study of E. coli food sensing behavior. Another outstanding example is the Shea-Ackers model of the bacteriophage lambda switch, which successfully models a biochemical switch using the equations of stastistical thermodynamics.
Those models most useful for tackling complex biological systems will probably be those models that in fact don’t get bogged down in a mess of complex details. They will be models built by stripping away complexity and by focusing on the much simpler core logic of the underlying, more simple processes that produce complex behavior.