[bip] BLAST and FASTA performance

[Andrew Dalke]

On Jul 22, 2008, at 1:56 AM, C. Titus Brown wrote:
> Barring a clear reason, why would biologists switch away from a
> successful approach?  Is SSEARCH significantly better than BLAST,  
> or is
> it just marginally better?

No fair!  That was my questions for you all, the practicing  
bioinformaticians. :)


The way I learned it, and what I've read, has been along the lines:

   "BLAST is faster but Smith-Waterman is more sensitive"

Searching for "Smith-Waterman sensitive" finds many quotes like that.


Ahh, here's a directly relevant news article on the topic
   http://www.genome-technology.com/issues/2_4/bruteforce/139714-1.html

    "But no version of Blast is comparable to the accuracy
     of the alignments provided by Smith-Waterman, he says."
       ...
    "Mueller thinks that Blast’s lack of sensitivity is a
     red herring when it comes to bench users who essentially
     see Blast as good enough."

and actual numbers (ahh, that makes me feel all warm and happy):
     "Though NCBI Blastp is still much faster, returning results in  
53.9 seconds, CLC’s Smith-Waterman completed the same search in five  
minutes — but with roughly 14 percent more hits than Blastp."


Didn't say they were good hits.  Perhaps it's the wine making me warm  
and happy?  :)

> "We used XX because it proved to be more sensitive in YY and ZZ cases
> than BLAST."

Which means you have to use BLAST *and* some other tool and compare  
the results.  More work for little gain.  And it's not like you could  
published that as an independent work because the original algorithm  
developers would have done that already.


> And yes, there's all the translational work needed to get a program  
> from
> the "algorithm works!" stage to the "heck, this is rock solid, well
> documented, and available in all of my important binary formats"
> stage...

People can have such silly requirements sometimes!


				Andrew
				dalke at dalkescientific.com