Scientific Service Industries

I was happily writing a reply to Jennifer’s neo-Ludditism, and it was becoming a whinge about the awful things biologists do when they aren’t doing their work properly (i.e. the way statisticians work). Then I started writing it as a blog post. I’ve now decided me eulogising Notepad (it even includes Comic Sans MS!) is not what’s required. And anyway, something more interesting came up.

The discussion over there drifted into the problems of getting support from bioinformaticians. I’m almost a bioinformatician – I’m a biostatistician, which is a bioinformatician who doesn’t have stupid amounts of data. Many of the comments are familiar to me, looking over the fence from the other side (admiring their grass). I’ll make my comments in the context of statistics, but I think most of them will transfer fairly well to bioinformatics. But, as with anything, your mileage may vary.
The frustration from the biologists’ side is nicely summed up by Richard Grant’s comment, which is worth quoting in full:

I’m carefully trying to avoid an ‘us and them’ thing here. I’m not entirely successful, but please keep my intent in mind as you read this ;)

My experience, having desperately wanted bioinformatical help, is that the bioinformaticians are too interested in proving the fossil record0, or something equally fatuous, to lend their skills to an interesting biological problem. Certain people from the group I’m thinking of have come all the way to Australia to give seminars, and people have walked away saying “What was the point of that?”

And before you bite me, Neil, that’s because what they were doing was biologically useless, not because we didn’t understand it.

I would love to have bioinformaticians on site, who wanted to collaborate. And I know such beasts exist – Neil himself has given me useful pointers (he’s 600 miles from me, so a more formal collaboration is not impossible, just more tiresome) and I appreciate that. But my experience is that Neil is unusual in this respect. (Having said that, my DPhil supervisor was a skilled BI and loves turning computers loose on interesting biology, too.)

Now I’m expecting that a lot of bioinformaticians will now crawl out of the silicon-work and say “Hey! I want to help you!” and that’s brilliant, really. But many of us more ‘wet’ biologists (including those of us who can hack a little bit of silico stuff) have been burned by uninterested comp biol people.

0 We know it’s true. What’s the point of saying this group of proteins is related to that group by this much, if there’s no functional information?

Richard’s view is one I come across a lot, and I think there are a few issues underlying it which are quite common.
The first is an institutional problem. At the places I’ve worked, the quality of statistical support for biologists has been pretty bad. I did my PhD in a large plant biology institute, and when I arrived there as a callow youth with just a Bachelor’s degree, I was the only person there with any statistical qualifications. This is hardly unique.
The unfortunate thing is that I chat to researchers and they say yes, they really need someone in their department to provide statistical support. But this message doesn’t filter up to the higher echelons. But hiring statisticians requires money and a commitment that can only really come from the top.
Some places do this better than others. For example, Scotland has BioSS, who provide statistical support to the biology community there. But in other places there is nothing except people with a bit of knowledge providing the support on a voluntary basis. The problem with this sort of support is that the people giving the support don’t always have the training themselves, so they rely on their own training they received from biologists, which means you see all sorts of odd pieces of advice (in the UK, professional training is available from the RSS, for those who want it).
Richard in his comment complains that when he has tried to get advice from bioinformaticians, they haven’t been interested, or they have been more interested in solving abtruse computational problems, rather than the biological problem that Richard is interested in. Again, I’ve seen this, and I have sympathy for both sides. The people he is going to for advice will not be paid to act as consultants (I hope not, anyway!), so for them the incentive comes from working on problems that are interesting to their field. Hence, they have rather different motivations. But this doesn’t help Richard – he might get his name on a paper, but doesn’t get his problem solved. How to get round this? Well, obviously if a collaboration were to continue successfully, both sides would need to find out how to make it work for both of them, and I will leave it to people wiser than me to suggest strategies.
There are a couple of alternatives. One is that Richard has to find somebody else to work with. Finding someone within his university or institution can be difficult for the reasons I’ve outline above. Working with someone who isn’t in the same region is possible, but (as Richard acknowledges) is not the same. Sometimes you just need to sit down together with a cup of coffee and a large pile of paper or a whiteboard and explain things to each other, or thrash out how you are going to deal with a particular problem.
The other alternative for someone having difficulties finding a good collaborator is that they educate themselves to do the work they need. Sometimes this can be done – the skills or knowledge may not be too difficult for a non-specialist to pick up. But it takes time, and wet-lab biologists are meant to be getting themselves wet in the lab. At some point there has to be a trade-off, when a biologists admits that they are a biologist, not a statistician or a computer scientist. A lot of the time they are also not going to understand the material as well as an expert could, and so are not going to do as good a job at solving their problems – a lot of time I’m asked questions that take me a couple of minutes to answer, but the questioner would only solve after a couple of day’s hard work.
So, what is a poor biologist to do? You tell me!

About rpg

Scientist, poet, gadfly
This entry was posted in Uncategorized. Bookmark the permalink.

10 Responses to Scientific Service Industries

  1. Maxine Clarke says:

    Interesting thoughts. (Does the cat agree?)
    A slight, but not complete tangent is what journals do. When we at Nature receive a paper with “considerable” (non-standard) stats in it, we ask a specialist statistician to peer-review it along with the more directly subject-related peer-reviewers. But even very basic stats, such as defining error bars, famously don’t get included in many submitted papers — and the odd one has even crept through the entire process to publication without anyone noticing. The Nature journals have created a statistical checklist for authors which is at our guide to authors (no 3f) and which we ask all authors to read at the revision stage of their manuscript. We receive practically no feedback about this list.
    It is surprising to me that very, very basic stats are often omitted or misused in submitted papers – by “basic” I mean saying whther an error bar is a standard deviation or standard error.

  2. Graham Steel says:

    Does the cat agree?
    Scottish tangent.
    My tropical fish hasten to add that this is only a partially unrelated comment.
    But in so far as far as “The Monster” and my musicy interests go, Bob O’H and cat may enjoy this offering from The Automatic. Yes, very 2007 but the track/vid is cool IMHO.

  3. Richard P. Grant says:

    Thoughtful post, Bob. Fancy moving to Sydney?

  4. Bob O'Hara says:

    Maxine – don’t get me started on refereeing. As you can imagine, I review quite a few papers for statistical content. The one thing I would say is that it can be rewarding to see a paper come back with the science improved (e.g. the results made clearer) because I’ve suggested using a better statistical approach.
    Graham – I enjoyed it. Jack is more of a metal-head. He also insisted that there is a difference between a monster and a beast, and he would never do anything monstrous. Actually, I think he doesn’t like being called The Beast, but doesn’t complain because he knows who cleans out the litter tray.
    Richard – I wouldn’t complain, but the reason the Beast (sorry, Jack) is with me is that he didn’t want to return to Melbourne.

  5. Jennifer Rohn says:

    I think every institute should have a dedicated bioinformatics facility where there is a small team of skilled, paid people who deal with the problems that the resident biologists need help with. Of the two institutes where I’ve worked that had such facilities, I had great experiences. This gets around the problem I frequently seen of running across bioinformatical problems that are just too time-consuming to compensate with a slice of blueberry pie, a middle co-athorship and a pat on the back.
    You get what you pay for.

  6. Richard P. Grant says:

    he didn’t want to return to Melbourne.
    I can understand that.

  7. Cameron Neylon says:

    What can you do? Shared and enforced coffee facilities. Does wonders for getting the right people talking together.
    Melbourne? Yes, I see that. But then if we had a choice to make we’d probably move back to Canberra, make of that what you will :)

  8. Richard P. Grant says:

    Canberra? Name rings a bell. Can’t for the life of me think why.

  9. Cameron Neylon says:

    Suburb of Wollongong I think. Never made it into your word of the week though…

  10. Richard P. Grant says:

    Oh yes, I think I’ve driven past it a couple of times…
    Neither Canberra nor Wollongong made it, but ‘Tuggeranong’ did. I should revisit that concept (word of the week, that is).