PFAM assignment, PROSITE motifs would be a good starting point in this case. You can try SUPERFAMILY which uses HMMER for known structures for any remote assignment, only if any in this case. Start playing with CD-HIT for a non-redundant set at 20%, so that you have a small set to look into. CD-HIT also provides clusters of similar sequences. This may help to look further.

Thanks. We are already using superfamilies from NCBI's CD-search tool. We've isolated about 100 mosaic proteins with the domain of interest (CLECTIN). I'll investigate more into the prosite motifs and CDHIT, as I'm not familiar with these.

Thanks

Where is the source of the protein sequences? Do you know where their corresponding genes occur within the urchin genome? If you do not, then you can consider using a local version of the exonerate program (http://www.ebi.ac.uk/~guy/exonerate/) to identify where each gene occurs. From there you can see if any orthologs have been previously identified in the Ensembl Metazoa database (http://metazoa.ensembl.org/Strongylocentrotus_purpuratus/Info/Index). If no pre-existing ortholog relationships have been identified, then you might consider doing a full genomic alignment between urchin and the genome of some other closely related, but well annotated, genome.

As an add-on, you can use just the sequence of the lectin domain as a query to search the genome of urchin, and potentially find more proteins containing that domain that you were unaware of.

Since there is not much conservation among sequences it appears the proteins are very distant sharing a few patches or block (segments .. whatever one may call). In Such cases using CLUSTALW2 as provided by EBI server will not be very effective. A more robust method/tool such as MUSCLE would be helpful. Also if the protein sequences of interest are transmembrane proteins then a bipartate method of aligning would be suggested. This can be done using PRALINE

I'd suggest HHPRED: http://toolkit.tuebingen.mpg.de/hhpred

You could also try CATH and Gene3D:

http://www.cathdb.info/

http://gene3d.biochem.ucl.ac.uk/Gene3D/

Thanks everyone.

Sequence similarity (by BLAST) must always be evaluated by approriate methods that validate the BLAST result. Respective algorithms leads COGs, SUPERFAMILY, Ensembl Families, EnsemblCompara, OrthoMCL, etc.

SYSTERS, http://systers.molgen.mpg.de/ , has clustered all protein sequences of year 2003 or 2007, mostly basing on the whole protein sequence space achieved from UniProt. Using this tool, you can work on sequence sub-sets that are maybe pre-classified in your sense. Find out, which SYSTERS families (clusters) are connected with your domain and take the larger sets:

http://systers.molgen.mpg.de/cgi-bin/nph-selectform.pl?select=PFAM

http://systers.molgen.mpg.de/cgi-bin/nph-selectform.pl?select=PFAM&v=5 (2007 data)

If you want to automatically retrieve data you can access data by WSDL web services

http://systers.molgen.mpg.de/cgi-bin/info_wsdl.pl .

If you need help don't hesitate to use the mailto link on the web site. Good luck!

Both are used worldwide

http://www.ebi.ac.uk/Tools/msa/muscle/ MUSCLE

http://www.ebi.ac.uk/Tools/msa/tcoffee/ T-Coffee

Dear Adam,

My suggestion is how you and are trying to identify, if possible, the function of these proteins. You use Blast2go is a pipeline complete http://www.blast2go.com/b2ghome,

because Blast2GO is an ALL in ONE tool for functional annotation of (novel) sequences and the analysis of annotation data.

- http://www.blast2go.com/b2gsupport/resources

- Yet, have one pipeline B2G4Pipe.

Try BCL:Align from the Meiler Lab at Vanderbilt. http://www.meilerlab.org/index.php/servers/show?s_id=1 ... it is customizable includes physical chemical properties and secondary structure prediction information. Used for distant sequence alignment and much superior to cursory BLAST searches.

Alignment can be done using any tools available. However, MUSCLE would do justice to the alignments. Try using BLAST EXPLORER available www.phylogeny.fr website.

You can select an option search nr/nt protein or pfam or other databases.

Once you aligned and had your consensus. In case poor alignment, you can search against the protein family databases to reveal the information you are looking for. Then search for the closest hit to get a pdb structure to work further on the secondaray structures.

Rather than using webservers you can Try SEAVIEW alignment software for MUSCLE alignments. Its user freindly - incase if you have any problems with the command mode operations of MUSCLE

Hope this helps

From my experience and others benchmarks, I would definitively drop ClustalW. It is the oldest aligner and not the best one. MUSCLE is pretty ok, but there are better tools that seems to work better. I would recommend three of them:

- MAFFT (with L-INS-i) options:

http://mafft.cbrc.jp/alignment/software/

- PRANK:

http://www.ebi.ac.uk/goldman-srv/prank/

- Clustal Omega:

http://www.clustal.org/omega/

Man, bioninformatics really needs to consolidate!

Bioinformatics was always considered as an enabling technology in the industry and Biology will consolidate Bioinformatics Tools as and when required.

I think Mafft is better, http://mafft.cbrc.jp/alignment/software/. If your sequence is very long, globalpair( G-INS-i) is better. If not, localpair(L-INS-i) is good.

I worked primarily with DNA and RNA. But, you need someone who can build machines, program in any language or tool and knows enough about molecular biology read books and tools searching for clues to performance. There are tools using FPGAs that will allow you to build a function specific processor. See "FPGAs: Instant Access" by Clive Maxwell for a high level overview. Code that uses the Intel SIMD instructions can require specific processors or expensive context switches. Nvidia and Intel make "accelerator boards" that require PCIe x16 slots.The latest have more features and better programming features. Nvidia compute boards have 5 GB or more (up to 12) of local GDDR5 and spawn sub-processes themselves. You can have up to 4 cards / machine exceeding 10,000 cores and work in the dozens of Tflops range. See "CUDA: Application Design and Development" by Rob Farber. I use the Intel Parallel Studio (with eclipse in Linux or Visual Studio in Windows) to generate better Intel native code. They come with several tools debugging, tuning, parallelizing, etc tools. Also, to program Nvidia Compute Board, the Nvidia Nsight is a library and tools that comes in a Visual Studio Version for Windows and an Eclipse version for Linux.

Expect more from your programmers, get them the training they need. You'll be doing everyone a favor.

[1] Pollard, Daniel A; Bergman, Casey M; Stoye, Jens; Celniker, Susan E; Eisen, Michael B (2004). "Benchmarking tools for the alignment of functional noncoding DNA". BMC Bioinformatics 5: 6. doi:10.1186/1471-2105-5-6. PMC 344529. PMID 14736341.: The authors discuss LAGAN, CHAOS, and Dialign as the most effective tools tested for certain uses.

Interpro might be usefull in your case. It gives you also GO classification that is more efficient than COG. Also you can try KAAS annotation, if your are lucky maybe you can reduce your list of unknown genes.

Conserning the multiple alignment based on sequence only, I would also go for MAFFT or MUSCLE. You will identify conserved segments, but it will not bring you to functional annotation. You can then use PHI-BLAST to search for similar protein with your pattern. In combinaison, you could run InterProScan on your sequences.

MUSCLE is one of the best options for your purpose according to my modest experience.