Are the inverse problems of Systems Biology impossible to solve?

Question

I have heard Sydney Brenner give a talk [0] on how the entire program of Systems Biology is suspect because, according to him, a chap named Hadamard showed that inverse problems are impossible to solve, or something to that effect.

I find it somewhat odd that no one seems to be paying any attention to this, and many are blithely carrying along trying to reverse-engineer biological organisms. I personally think that this reverse-engineering effort has its place, but Sydney Brenner doesn't seem to think so, and advocates tackling only the forward problem. Here's an excerpt from an abstract to a similar talk of his:

... This notion of computation is, in my opinion, the only valid approach to biological complexity and is opposed to many of the ideas underlying what has come to be called systems biology, which is very fashionable today. It will be shown that systems biology attempts to solve inverse problems — that is, obtain models of biological systems from observations of their behaviour — whereas, what I call computational biology, continues in the classical mode of discovering the machinery of the system and computing behaviour, solving a forward problem.

Is Brenner simply wrong on this or is there something to his objections?

---

[0] Sydney Brenner, Much ado about nothing: systems biology and inverse problems, 2009.

Answer 1

The answer is mostly no, and this is not as disruptive a statement as it seems to be.

I think Brenner has a point of course and he is not the only one who wonders whether the systems biology can create a holistic model of living things. Still I wouldn't take this to mean we should stop doing systems biology research.

At 18:30 brenner illustrates the inverse problem with some examples 1) can you create a drum from only the sounds that come from the drum? (no he says). 2) can you solve the molecular structure of a molecule from a diffraction problem (no he says).

Of course we can do both of these, but the point is valid as we need to have an idea of what a drum looks like. If we did have even a glance at a drum, this lets us build a model (cylinder of diameter d, height h and with a drum head of a tension t and material properties alpha beta, etc) from the sounds is a lot of work, but it could be done.

The problem of solving a molecular structure from diffraction intensities alone has also been solved for small molecules in the 60s I believe by so-called 'direct methods' which don't work well or proteins and larger molecules. Still the point is that you had to have a good mathematical model of molecules and diffraction to get to direct methods, MIR, MAD and other diffraction only methods. Sorry I can't go into details here...it might bore you to death if you don't care about crystallography.

In both cases the statement is that you need to have that model - what does a molecule look like, what does an x-ray do when it encounters a crystal? In order to get there. Which makes total sense and I agree with.

I wouldn't read too much into this though, because in practice this has never been done, yet Bragg started with just the diffraction pattern and drew in models from other fields of study. Given the nature of the proof described, no scientific discovery has ever happened in a mathematical vacuum and systems biology will not either.

Rather I would read this talk as saying that the models we are using for action are not very good yet. What is happening is that synthetic biology (engineering approaches), experimental observation, and intensive modeling (the 'inverse' and 'forward' approaches Brenner mentions) are all happening at once and are informing each other.

Its hard to imagine that data from inverse approaches can't be an important part of any solution to biology. Its also looking likelier that a mathematical model alone of biology will be difficult or impossible to find also. What is currently (IMHO) working is a framework that is both conceptual (and not entirely quantitative) as well as analytical (mathematical) and data driven.

Answer 2

The article is well known and discussed amongst systems biologists. It makes a good point - high throughput, observational biology cannot substitute for mechanistic studies that provide causal information.

But we know an awful lot about biological systems already. All systems biologists are using information from the forward approach implicitly when they formulate their models. More importantly however, most systems biologists combine the forward and inverse approach in their work. Models are tested and validated by perturbing the systems, and updated according to the results.

Discussions of these kind often devolve into a debate over the semantics of systems biology, in which it's detractors attempt to define "systems biology" in the most unflattering terms possible.