Bioinformaticians, take heart

It’s difficult for bioinformaticians to publish in so-called “high impact” journals at all, never mind as first author. Many of us are not group leaders with tightly-defined research programs; we are the “go to” guys, happy to apply our skills to any dataset that comes our way. In academia at least, we’re caught somewhere between research scientist and IT support. It can be a frustrating life.

So it’s good to see that Nature, a journal not renowned for publishing articles with a strong computational biology component, has seen fit to publish this:

Structure-based activity prediction for an enzyme of unknown function
Hermann, JC et al. (2007)
Nature 448: 775-779.
Abstract | Full text | N & V

It’s a fascinating piece of work. The authors started with an enzyme of unknown activity from Thermotoga maritima, a thermophilic bacterium. Initial structure/sequence analysis suggested a superfamily for the enzyme. They then compiled a list of ~ 4 000 potential substrates, modelled tetrahedral intermediates that could resemble the transition state for each one and performed docking simulations of each model with a model of the enzyme active site. This generated 4 candidate substrates, 3 of which were confirmed by biochemical assay. For a finale, they determined a crystal structure for the enzyme + bound product which agreed closely with the model and identified a new metabolic pathway for orthologues of the enzyme in other genomes.

There are a few criticisms of the approach. This is a case where modelling was a good approximation to reality, but there are plenty of cases where that isn’t true. It also relies on quite a lot of prior knowledge concerning sequence/structure and metabolism, which isn’t available for many uncharacterised proteins. And it’s currently quite computationally expensive, so not exactly high-throughput, although that will change of course. Still, if you enjoy genome-scale bioinformatics and structural biology, it’s almost enough to make you drool.

Technorati Tags: , , ,

5 thoughts on “Bioinformaticians, take heart

  1. There was a similar paper by Matt Jacobsen and co-workers (either in BMC or a PLoS journal) published a month or so ago. That said, this is quite impressive, especially the bit about transition states. Shoichet is quite a heavyweight in the molecular modeling world, which probably helped get this into Nature (plus the fact that they actually got a crystal structure as well). Even a couple of years ago, there was no way something like this would get published in nature..

    The cool part … this validates a lot of what we were doing at my first company over 5 years ago.

  2. Without having read this paper very carfully, I have the impression that the authors could have found approximately the same result by just running their sequence against the Pfam database (without the Nature paper, that is)

  3. A Pfam search would certainly classify the sequence as an amidohydrolase (I just tried it myself). However, that only leaves you with a computational prediction. The paper might be overkill in some respects, but experimental validation of bioinformatic analysis is always good to see.

  4. Depends on the kind of granularity you want. At the level of “amidohydrolase” Pfam might do the trick, but at the level of identifying the unique combination of sequence, structure and lead family, this does have some unique information content.

Comments are closed.