Search This Blog

Monday 11 July 2011

Automated Protein Structure prediction... or probably not.

I was browsing through a bioinformatics textbook last week, and read up about the work of Prof  David Baker in the field of protein structure prediction. One of the resources that the Baker Lab offer for free to academic customers is access to their Rosetta methodology via their Robetta structure prediction software hosted on their own computers. It is a really easy to use resource with minimal hassle in terms of registration. Anyway I thought it might be interesting to see what an automated program might be able to make of the single protein NHase of Monosiga brevicollis which is made up of the standard two separate alpha and beta subunit proteins tied together, both of which are well known from 41 (and counting) fairly similar PDB files which have a high degree of similarity. First of all, you submit your FASTA sequence, and it parses it using the Ginzu domain prediction technique.
I was pretty pleased with this start as it spotted the two domains by BLAST, and match them to two PDB files of cobalt containing NHases. I then set it away to do the full structure prediction thinking that knowing those two starting points, most of the heavy lifting in terms of the structure prediction had been done. The prediction said it would take seven days but actually was finished after two (not a busy weekend for their computers it would appear!). Its output is the five best structures...
I have to say I have no confidence in any of those structures... all look a bit long and dangly for me. If you contrast them with a couple of PDB structures of NHases of the usual type below you can see that they tend to be much more compact- look where the two subunits' beta sheets are in the PDBs and in the prediction.

I guess it was asking a lot to get a reasonable guess at the histidine-heavy chain in the middle of the Monosiga NHase, and that is a crucial part of this prediction. No need to welcome our robot overlords just yet, and I need to go and talk to a human homology modeller!

No comments: