Developing statistical and computational methods is fun…
Figuring out the mathematics, making the algorithms more efficient, using every programming trick in the book — it’s like producing a piece of art. The rush of adrenaline, when you discover a bug; the satisfaction, when it all comes together and does what it is supposed to. Faster and more accurate, hopefully, than all competitors. Victory!
I can completely understand why so many people in computational biology are busy developing more and more methods. I even got some of my own. It’s fun and an intellectual challenge – the scientific equivalent of cross-word puzzles.
… but computational biology has too many methods already
But more and more I find myself wondering if I should indulge in this fun exercise. Computational biology has too many methods already. Only a tiny number make an impact. Most are just marginal improvements to existing methods. Not what you would call a game-changer.
There are some exceptions of course. Every time a new technology comes out, it’s like a gold-rush for the methods people (like right now with all that sequencing going on). And that’s exactly how it should be. But after a while, there are no new ideas anymore and everybody seems to be doing more or less the same thing. That’s when we are back to having too many methods.
Novel insights trump novel methods
But what we need more than methods, are insights. In general novel insights trump novel methods.
That’s true, even if you are a hard core methods person from a comp sci background: what better way to show how cool your method is than proving that it drives science forward. Claims of novelty (‘Our approach is the first to..’) are less useful than claims of insight (‘Our approach showed that..’). In the first case you get into a fight about who was there first, in the second case you talk about science.
Just to make myself clear: I’m not a luddite; my rant is against incremental advances, not against methodological innovation per se. I really like thinking about algorithms and stastistics and I like it when things are done right and not with a crappy heuristic. What I don’t like is ‘They (the biologists) do the questions and We (the computationalists) do the methods‘. That’s wrong. Our computational work must be question-driven, too.
Ok, enough whining – are there any ways out of this situation? I can see two options:
- If you are a computational biologist, aim for uniqueness.
Try to find a niche and solve problems that haven’t been solved thousands of times before. If you stay close to your biological collaboration partners, there is a chance that you spot a new problem, one that you can be the first to solve! That’s even more fun than improving on 10 previous approaches. Don’t get into a field just because everybody else is doing it. Only get started, if you think you can contribute something unique.
- If you are a computational biologist, aim for big questions.
Make sure you contribute to some bigger biological/medical question that your collaboration parterns are interested in, too. Sometimes there is no need at all to develop something new; an old and established method can do the trick perfectly. Focussing to hard on methodological novelty can prevent you from insights a seemingly boring method can provide.
Most of us will be finding themselves rocking back and forth between these two positions. Right now I’m putting more emphasis on the biological bit (which is not very surprising given that I work at a biomedical research institute).
My next post will discuss three positive examples of computational analyses with big medical impact.
This is the first post in the series. Once more come out, you can find all of them under the tag ‘method versus insight‘.