If you are looking for evolutionary distant proteins , you can try the PSI-BLAST and modify the default settings after each iteration !!

Exclude those taxon and keep one species in, u can keep out others in your result

Are you already using only the refseqs as database to search?

I do not suppose since I just used the default setup of Blast.

As I see that refseqs is non redundant database, probably it will be

more "clean" the default database (even if probably still hosts

sequences coming from very close organisms).

Again I wonder if there is a simple cut-off parameter in BLAST to

exclude from results sequences too similar.

This guy has some words to spend about it, I guess :)

http://www.youtube.com/watch?v=Ud_6VpX5AgI

I.e. Try with the refseq database and the blosum45 matrix for comparison

Hi Gianluca, As others suggested you can use ref sequence databases particularly with reference to your sequence. MEGABLAST using high E-value thresold can also minimize the number of sequences. However during BLAST searches paralogs cannot be eliminated completely in case if you are trying to construct a phylogenetic tree use highly identical sequences and you should include the sequences from different type strains (Due to the variations in the genes within the species). Once you have set up the thresold E-value and select only those sequences you are interested in to get the FASTA files.

Good Luck

Cheers

Aditya

Thank you all for your comments. I'll try in the upcoming weeks suggestions.

Gianluca

What you can do is to take all your results with BLAST, and then pass your unaligned sequences to CD-HIT:

http://www.bioinformatics.org/cd-hit/

Nice tool to reduce redundancy in a bunch of sequences.

Hi Gianluca,

is it not an advantage to have so much alternatives, is it? The question is: where does the alternatives come from? Are you interested in a particular reference sequence and you do not want to mitt it? UniProt includes much identical or fragmentary sequences into the database, so, it would be another advantage to reduce the basic set. UniRef (90, 50) would here be an solution. But what you find in a respective subset is very inhomogenuos. Take care about this!

In SYSTERS ( http://systers.molgen.mpg.de ) we reduced the full protein sequence set (2003; across all species; 1.2 Mio protein sequences to ~600 k sequences by a 80/80 criterion: 80 % similarity in 80 % of the sequence length) towards the so called non-redundant sequence set. If you do a BLAST there against the SYSTERS consensus database (~150 k clusters) you will meet the protein family (with multiple non-redundant sequences but also with all redundant sequences, too). The question is: Do you have new sequences and do you want to find the meaning (functionality) ? Which is the reference sequence you are looking for? Maybe the optimal strategy is that you meet the family easily and quickly and then go into the detail with all similar sequences in/of that family to find the most applicable feature as the other people were mentioning (species, paralog-ortholog, gene functionality, etc). So have a look into SYSTERS or any other protein family approach.

Greetings, Thomas

Hmm. If I am right u r talking of getting less than 10 similar output per query in ncbi BLAST (Which is min option), and in that case try BLAST2GO

Have you try with PSI-BLAST?

I just noticed this post which may provide you with some helpful methods also:

http://www.researchgate.net/topic/Bioinformatics_and_Computational_Biology/post/Looking_for_a_method_to_filter_out_data_from_related_BLAST_results2

We did a prototype system that does this: http://www.ogic.ca/projects/bluster/

Sorry not understand alot about molecular and gene sequences.

In the BLAST search form, in the Choose Search Set box, there is a line for ORGANISM with a check-box for EXCLUDE next to it. This can be very useful for the problem you describe. Using the Taxonomy browser:

http://www.ncbi.nlm.nih.gov/taxonomy/ you can fill in the ORGANISM with either the name or tax-ID number at any level of the taxonomy you wish, and then ask to exclude hits from those taxa.

For example, if I am interested in mammals and I have a Mus musculus gene of interest that has been sequenced from thousands of strains of mice, I can exclude all Mus musculus, or all Mus genus, or all rodents, in order to just find the gene in other more distantly related species. I usually also do a second run, exluding all mammals, to get a marsupial, fish, amphibian or other sequence(s) to use as the outgroup, if I am interested in the mammals.

You may want to use JALVIEW (http://www.jalview.org/). It is ideal for your problem.

1. Download your sequences raw or in an alignment (many formats are recognized).

2. Load them into Jalview, or you may use JalView to download the sequences through their accesion numbers (File fetch). Many databases are supported.

3. If not aligned, use Jalview abilities to request alignments from external sites: MAFFt, Probcons, clustalX, T-coffe, and muscle are supported.

4. Use the Edit/Remove redundancy tool te cut down sequences that are too similar to each other. Just select the cut-off value (as you move the bar the programs highlights the sequences to be removed, so you can easily select the desired level of similarity to use as cut-off).

5. Save your file. Again many formats are available & voilà - you are done.

good luck

Try cdhit