nodalpoint.org

Wolbachia genomes sure was a conversation stopper (pauses as tumbleweed blows across the screen)...perhaps I can share a few more thoughts on getting things done. One of my favourite topics, as an ex-wet lab biochemist turned computational biologist is the differences in working practice and mentality between these two types of scientist.

In our last discussion, we considered whether research is "non-linear" and I think that this notion is somewhat illusory. We do have goals to achieve and we take a stepwise path to those goals, but the steps may branch and are constantly being revised. In other words - we're making it up as we're going along.

I think a major difference between wet lab versus computational work is the degree to which we make it up as we go along. Here's a couple of examples. In the lab, you want to purify some RNA, reverse-transcribe it, end-label it and use it as a probe. In all likelihood, this has been done many times before and standard protocols have been written down which describe this task. There may be a few protocols with minor differences and you may experiment with slight alterations which improve the procedure but essentially, you have a recipe to follow. Mastering this recipe is a lab skill and generally, if you fail, it's because you did something wrong or one of the components was sub-standard.

Now a bioinformatics task - you want to take all the ORFs from a genome, BLAST them against the subset of sequences from the PDB, extract those queries that have a certain identity to a hit, thread them against a set of templates and build a homology model of the query protein. The difference here is that whilst you still have several steps in your procedure and some idea of what "protocols" (i.e. software) to use at each step, there are multiple ways to achieve this task. Do you want to grab the query sequences once from an online database (wget/ftp/an rsync server/a Bioperl fetch tool) or regularly obtain them for storage on your hard drive (mirror)? Once obtained, would you like to index them for easy retrieval? Would you like to speed up the BLAST search using a parallel implementation? How do you decide on a suitable cutoff above which query/template are suitable? What tool do you use to parse the BLAST report? What choice of threading software? Are you happy with the pre-supplied template database or do you want to build custom templates? What's the output format from threading - plain text, marked-up text? If the latter, how about tools to transform it for database storage or display? How about your final model PDB files? Do you want to check their quality in some way, view them graphically, how are you going to analyse them? And handle the output of the analysis tools? Is this a one-time project or something you'd like to automate as a pipeline and run regularly? How will you share the results with colleagues?

My point is that in computational work, we are actually making up the methods as we are going along and always considering new and better ways to perform tasks and achieve goals. Perl people like to say " there's more than one way to do it" and I think in computational work, there are multiple ways to do pretty much everything. Herein perhaps lies the biggest communication problem between wet lab biologists and computational biologists - I think the former group are frequently taught that methods are no more than a means to an end - it's the result that's all important. In computational biology, we spend a lot more time thinking about methods, because choice of method has enormous bearing on the quality of the result, the speed with which we obtain it and the ways in which we can further use it.