Blast against viral database vs NR

[arundurvasula]

Hi,

I'm trying to blast(x) some contigs I've created. The problem is I have a ton of them so I'd like to reduce the search space by using a database of only viral sequences. I created a database like this, but the results seem off. For example, when I get a sequence that matches with a plant virus, taking that sequence and using NCBI's web blast interface against NR returns plant sequences.

My thinking is that blast returns whatever sequences the query is close to, and when checking against viruses there are only viral sequences so it returns whatever viral sequences there are. When I take the sequence to NR, there are many plant sequences that are much closer so it returns those. Hopefully that makes sense.

The main problem with this is that I can't trust my results. If I get a virus that I'm interested in after searching the viral database, I have to use NR to make sure that it's actually correct. So my question is, can I do anything to make sure when I'm searching against the viral database my results are actually accurate?

[mbblack]

BLAST inherently will always give you a best match against whichever database you query. That's how the algorithm works. And things like bit scores and e values are dependent on the size of the database searched, so are not comparable between databases.

If you only search against virus sequence, you will only get a virus as your best possible match. If you suspect plant contamination in your sequences, then you need to include both viral and plant to make any judgements about homology match.

Homology is not a yes or no outcome. It is a judgement call based upon the similarity of the alignments you get back. Searching against only viral sequence when you know or suspect you have non-viral query sequences will inherently bias your interpretation of what you have.

So I would say you should be using the NR database, regardless of the overhead that adds to your computations. Homology determination by sequence similarity and alignment is entirely dependent on what you query your data against. Using only viral sequences would only make sense if you had a strong basis for assuming your search sequences were uncontaminated by anything other than virus.

P.S. your results against the viral-only database you made will actually be accurate, but they will be accurate for that specific query and reference set. Similarly, you results against NR will also be accurate and correct. BLAST and any homology matching are context dependent, always.

[GenoMax]

What is not clear in the original post is if the sequence set is only supposed to have viral sequences. So if that is true then using the smallest database should make the search more sensitive.

[Richard Finney]

To complicate things, don't viruses sometimes have genes which are homologous to their hosts genes?