Here is a cheeky post along the lines of Can a biologist fix a radio? (well worth a read). The discovery that Pluto wasn’t the size we expected got me thinking about how we would have tackled it. So in our case: *Can a computational biologist do astronomy?*

Our protagonists are a Mathematical / Computational Biologist **MCB**, and a Noteworthy Astronomy Scholar with Antiquated ideas, **NASA**. (If anyone’s initials are MCB, I forgot and genuinely didn’t mean you!)

**MCB:** I am going to make a model of how Pluto moves under gravity.

**NASA:** we know that Pluto is subject to gravitational forces, so its orbit depends on its total mass according to Newton’s law of universal gravitation.

**MCB:** Great, so if I could measure the density and radius of Pluto then I could work out its mass with M = density x volume, and then I could work out the gravitational forces on it.

**NASA:** Well, if we say it is a uniform sphere, then I suppose so yes.

**MCB:** Good. I will do some calculations…

**MCB:** I have varied the radius and density of Pluto and discovered that Pluto can have almost any orbit you like.

**NASA:** By the way, we have watched the orbit, and it is very consistent.

**MCB:** Interesting! I will do some more calculations to see which densities and radii it uses.

**NASA:** Er, we don’t really have any evidence that Pluto is varying its radius and density. It might make sense just to work with its total mass for now.

**MCB:** That’s all very well, but we don’t have any evidence that it isn’t changing its radius and density all the time do we?

**NASA:** Er…

**MCB:** In fact this would make sense, because Pluto’s orbit must be robust to fluctuations in radius and density that occur when space dust hits it stochastically. It probably varies them continuously to adapt to getting hotter when it is closer to the sun too, and all sorts of other things. I will do some more calculations…

**MCB:** The radius and density of Pluto could take almost any values and still give that orbit! Pluto is robust to changes in its radius and density [as long as they fall near a line that looks a bit like 1/density is proportional to radius^3].

**NASA:** [You’ll probably find that constant of proportionality is 4/3 pi], I think we might need new equipment to measure the radius and density, you should probably stick with a simple model with just total mass for now.

**MCB:** All the evidence we have is completely consistent with Pluto taking a wide range of radii and densities, I can make some interesting predictions for what happens to Pluto in new situations at different radii and densities and publish lots of papers about them.

**NASA:** Well it might take various values, but we’d better go and check.

[Some time later…]

**NASA:** We’ve just measured Pluto, its radius and density weren’t what we guessed, but they take constant values!

[The End]

In case you didn’t get my not-so-subtle point, I have read quite a few papers linking un-identifiability of parameters in inverse problems as plausibility that the parameters really do take a wide range of values, backed up with an argument that they must to confer biological robustness, or explain variability. Biological systems obviously do have to be robust to stochastic transcription, noise and a wide range of conditions. But lack of information about a quantity is not evidence that it varies in reality: certain things are very carefully controlled by numerous feedback mechanisms e.g. blood temperature. Don’t get me wrong, I think understanding variability and uncertainty is absolutely crucial. But if you are relying on something varying a lot for your results, I would say the onus is on you to design an experiment where you *can* measure it as well as possible, and do it lots of times to show *how* it varies.

Pingback: Numerical errors from ODE solvers can mess up optimisation and inference very easily | Mathematical Matters of the Heart