We are all excited by the experimental observations of gravitational waves. 100 years after Einstein published his General Theory of Relativity, one of its most important predictions has been experimentally validated.
General Relativity is, of course, one of the great masterpieces of theoretical science. And the experimental validation itself is also monumental. Most of us theoreticians are not, and cannot hope to be, as great as Einstein. But as someone who works in theoretical (or computational) biology – as opposed to experimental research – it is helpful to reflect on what makes good theoretical work, even if my contributions are on a much smaller scale.
Thankfully, Karl Popper, another great 20th century mind, has already helped us with this: good theoretical work makes predictions that can be tested experimentally, to be falsified or validated. Moreover, the more general the theory, i.e. one that can make a wider range of predictions, the better. Popper trained as a physicist, and this works extremely well in Physics, but does run into some problems in Biology. There is one way in which Biology is ahead of Physics: it has a “grand unifying theory” that Physicists don’t yet have. Organisms evolve by natural selection. However, while this theory can make clear predictions (e.g. my colleague Pete Lund had students evolve acid-resistant strains of E. coli by successively growing them in lower and lower pH), the predictions are at an organism level – it cannot predict exactly how it will evolve, or the molecular mechanisms.
Other theories can be smaller. Several of our own papers have made the point that negative self-regulation facilitates gene expression at lower metabolic cost to cells. (Rival theories point to reduced noise, or increased response time). This is applicable to a wide range of genes, in a wide range of organisms. It is likely that all of the theories are right sometimes and irrelevant sometimes: Biology is messy like that. But the theories are still experimentally testable and falsifiable.
And theories can be smaller still. One of our most recent papers predicts that the expression of the ZntA protein in E. coli should, under “normal” conditions (exponential growth in LB is of course entirely normal!), be heterogeneous in a population of cells. Very specific! But also very clearly testable and falsifiable.
So now to my two frustrations (and the point of this blog post). Theoreticians who do not make falsifiable predictions, and experimentalists who think that our predictions have no value without experimental testing.
Theoreticians should make predictions. I distinctly remember during my PhD one well respected theoretician proudly saying that no experiment could falsify their model. What is the point of their model then? Likewise, I have seen theoreticians tell experimentalists that their data must be wrong because it contradicts their model. No – just don’t go there. Good theoreticians must always respect good empirical data. And I read plenty of models of molecular or cellular systems that are all very nice, but don’t actually make any predictions about the system (although of course many do). There is, of course, an exception: sometimes computational biology can be very useful as an experimental system in of itself, to test hypotheses that cannot be tested in the lab or field. For example, simulations of evolution can allow testing of hypotheses not possible in a lab (e.g. work done by my first PhD student Dafyd Jenkins). But even then the models should make predictions – even if they are very difficult to test in vivo. We predicted that molecular mechanisms for stress response should be more ancient than mechanisms for optimizing substrate utilization: very hard to test, but still a clear prediction.
Equally frustrating are the experimentalists who do not appreciate that our work is meant to finish with predictions. I have heard this one so many times. Tw0 examples from our own work: Chris Fernando’s remarkable model showing that single cells could carry out associative learning – an experimentalist thought that it shouldn’t be published unless we confirmed it experimentally. What nonsense! We are theoreticians – it is our job to come up with ideas and predictions – it is for others to pick up the baton and try to build such a cell, just as Einstein can predict gravitational waves and leave it to others to try to find them.
We had a similar problem with Hiroki Takahashi’s zinc modelling paper. We finished with the hypothesis that ZntA should be heterogeneously expressed. The referees from the first journal we sent it to all said that we needed to do the definitive experiment for the paper to be published. They completely miss the point. The fact that we end with a clear prediction is a strength of our paper, not a weakness. It shows that our modelling work (and indeed experimental work – the paper has experiments too) was successful in being able to come up with a clearly falsifiable idea.
Back again to Einstein. What a triumph. To have a theory that is so general that it pervades all mass on Earth, yet makes predictions that are very hard to test, so hard indeed that 100 years later, it took a team of hundreds of scientists to confirm.