Back to roots
I like bioinformatics because it is so useful at pointing out the next useful experiments and helping to extract knowledge from your data. This is why I think it is possible and useful to do experimental work alongside with computational work.
I have spent the last week back in the bench doing some biochemistry. I usually don't do much bench work although I have a biochemistry degree. It is at the moment not easy to keep up doing my computational work while doing the lab work but I want, until the end of my PhD, to find a way to keep doing both things at the same time. I should divide my time between the two mind sets but I am not sure of the best way.
Any ideas ?
Saturday, December 10, 2005
Posted by Pedro Beltrão at 6:19 PM
5 comments:
I'm also a biochemist by training and now a computational biologist. Unlike you though, I don't miss the benchwork very much.
My situation is probably different to yours in that where I work, there's an awful lot of bioinformatics to be done (we work on several genomes) and not many people to do it (myself, basically). So someone had to make a decision to go 100% computational, which I was happy to do.
I imagine that balancing bench and computer work is very difficult. You don't want to be in a situation where you are continually running back and forth between the two. It's possible to be very productive in a few hours computationally, provided you can concentrate fully for that time. On the other hand, benchwork takes "as long as it takes" due to the time constraints of protocols (waiting for cells to grow up or whatever).
Ultimately I suppose it depends on the exact nature of your work and your own style of working. If you're very disciplined and have good time management skills it may work for you, but I wouldn't like to try it! Perhaps others will share their experiences.
Well, this is the exact issue that I face these days...
I've done mostly experimental work in my career, but always with a lot of software development just to handle and analyse the data. Now, I've been doing 98% computational genomics work, with a tiny bit of bench work here and there...
maybe some questions:
(1) Is there enough time to do computational and bench work at a fast-enough rate to publish? This is really field and area-dependent.
(2) Career: Data-monkey vs. original concepts. If you are developing original concepts/algorithms, then straight computation is fine for the future. But what if you mostly (like me) end up doing large amounts of "data handling", which is nontrivial but not hard?
(3) If you are not creating great new conceptual frameworks for analyzing data, then... do you need benchwork to move forward (similar to 2, above)?
***Put this another way: is it good enough to do original, interesting analyses of data using existing tools (sort of the way bench science works - most papers are original, exciting studies using existing techniques - few papers include technical innovations...)? Or does moving forward in bioinformatics require new concept development?
As for me, I'm naturally good at handling large sets of data and, yes, I can think statistically, but I am in no way a real statistician or someone developing novel approaches (with one or two exceptions) - so this seems to be pushing me toward the bench, again...
Feel free to comment or email (see below for email address)
-Mark
markb7002000
(at) yahoo
(dot) com
It's fascinating to see people having the same issues with their careers the world over!
My comments to your questions/points, not necessarily in order:
(1) A lot of what I do can also be characterised as non-trivial but straightforward data handling. Don't discount it, it's a skill that more biologists should have. There is plenty of room in bioinformatics for what I call "intelligent use of pre-existing tools" (e.g. data analysis pipelines using Perl wrappers to applications), as opposed to algorithm development. In my opinion, far too much of the current bioinformatics literature is focused on tool development as opposed to useful applications and discoveries using the tools. I mean, how many ways can there be to align two sequences?
(2) If you want to do purely computational biology - that is software development, new algorithms or hardcore mathematical/statistical procedures, then you want to be working in a purely computational environment with like-minded people. You will never have the time to focus properly on this kind of work if your role is essentially one of IT support with biologists clamouring for your attention.
(3) When do you become "real" (as in real statistician) ? Most of us are self-taught. As someone who can handle large data sets, think statistically and develop software, you sound streets ahead of the average biologist. I had a similar lack of confidence when I first started programming - when would I be a "real" programmer? Confidence comes with time and experience. Ewan Birney at the EBI once said something like "don't worry if you feel stupid when you first start, we all do".
The important thing is to get yourself into a work environment that suits you. If you're happy doing some benchwork and also playing the role of "lab computer guy", that's fine. If you feel that you are moving in a more purely computational direction, such a role will ultimately frustrate you and you'll want to be in a purely computational biology research setting.
I relate to the question "when do you become 'real'?" I don't particularly want to be identified as a statistician (and there is no danger of it). But I would like to acquire enough vocabulary and an appropriate mindset in which I can at least formulate my questions. Too often, we dissuade people from other fields by dismissive comments (it is the case of my "biostat" colleagues). So what if one dabbles? it's still learning after all.
It is true that it really depends on the field. In my case I am mostly doing protein interaction prediction and although I can use benchmarking to show the quality of the predictions, there is a lot more impact in showing some experimental evidence for new interactions. I think this is also due to some distrust in the numbers :). Usually researchers tend to believe more in a blot than in a benchmark with gold standards.
Another important factor is the limitations of preexisting data. Sometimes your analysis can lead you to an interesting question that you just don't have data to use to answer. In both cases it might be probably better to try to find collaborators. Anyway, I will keep on trying.
Post a Comment