Friday, 5 December 2014

From the trenches of experimental biology

I remember a conversation I had in undergrad with my two best friends, both of whom where studying physics. We were discussing the relationship between scientific equations and the actual natural world over cups of tea (as you do when you're a bunch of over-achieving Cambridge students). As physicists, they were of the opinion that the equations had meaning, and that error was just that: the result of experimental imperfection. I was unconvinced: I could not perceive of any situation in which all biological observations would line up neatly with the predicted curve. There would always be error. Here lies a big division between the physical sciences and biology: we cannot get away from variation in biology. It it structured at the most fundamental level into the data we work with.
And sometimes, that variation is the dominant signal in your data. And that variation is both biologically relevant and analytically intractable. Which brings me, quite neatly, to my data. I'm currently working on a poster for the upcoming SICB meeting in West Palm Beach this January (conference location win, incidentally).  I'm looking at whether or not our experimental treatment is related to changes in how the tongue and jaw move in feeding. Our preliminary results (and these were reasonably robust preliminary results) showed a clear pattern. And then we added more sources of variation.
There was a point earlier this week where I feared the whole study would collapse, and that the interesting result was simply an artefact. As it turns out, elements of that pattern are consistent, but the variation between individuals and even between feeding bouts, is huge. Yet interestingly, within an individual and within a feeding bout patterns are highly consistent. The variation is non random. The variation is data.
The whole point of this project is to attempt to understand a system that is both highly integrated functionally and anatomically, and yet also highly variable. What's more, clinical prognosis for injury to the part of the system we're studying right now is also highly variable: symptoms of varying severity, duration and response to treatment efforts. We do work in a animal model precisely so that we can create reproducible, controlled injuries to the system. These are ceteris paribus (to use my PI's favorite expression) experiments. And, all things being equal, the system responds differently. This might lead some to look for some other systematic difference, and maybe there are some (stage of neural maturation, variation in anatomical patterns innervation, changes in spontaneous brain activity, or simply different strategies to adapt to the perturbation are all avenues we're looking at). And yet, at the base of it, I don't think there is any equation that we could make, even in an ideal world, that would result in our data lying exactly on the predicted line. Because variation in biology isn't noise, it's signal.