Re: [modeller_usage] Searching the PDB for sequence with structure

[Jacob Glanville <>]

Hi florian,
Try sending your sequences to the phylogenomics.Berkeley.edu  
webserver: it has a structure search component. I was involved in  
building it, and can vouch for the pdb database we built to be drawn  
directly from the Pdb files: only residues with coordinates are used.  
Also, the search uses hidden Markov model alignments, which will  
generally outperform blast when aligning loops with more remote  
homology. They might even still have a batch submit option....

Let me know if you are still having problems: I work with antibodies a  
lot these days, and consequently have a number of techniques for  
mining short loops out of Pdb that you might find useful.

Jacob Glanville

On May 28, 2008, at 4:12 AM, Florian Odronitz <> wrote:

> Hi Christian,
>
> > If no one will come up with a better solution, you might want to  
> > build
> > this tool yourself:
> That sounds like fun!
> > All you need to do is to download the PDB files and check whether the
> > sequences in question are contained.
> The thing is that the sequence might be there but with gaps. It would
> be nice to have that be matched by BLAST, since regular expressions
> are not suited for this.
> If I am not the only one who has this problem I might do something  
> like:
> - Download all PDB Structures.
> - Extract the sequences of the proteins as well as the sequences that
> have ATOM records.
> - Make a quick alignment between each of those two sequences.
> - Put the two sequences and the PDB ID in a database.
> - Make a BLAST database file witch the full sequences.
> - Make BLAST accessible on a web server.
>
> The search results could be presented as an alignment of the sequence
> that was searched for and the sequence that actually has structure.
> It would be easy to implement queries like "search for sequence X and
> return only results where more than 30% has structure".
>
> > checks) is contained in the biopython project ( http://
> > biopython.org/ ).
> I would prefer to do it in ruby, but same same.
>
> > MD is not an option for you?
> I think the stretches are too long (several dozen aas) and to numerous
> (~100 structures * several gabs/structure). I would also feel better
> if I could use experimental data.
>
> Cheers,
> Florian
> _______________________________________________
> modeller_usage mailing list
>
> https://salilab.org/mailman/listinfo/modeller_usage