galt,

On Jul 14, 2009, at 12:12 PM, Galt Barber wrote:

> Looking at the source code for gfServer, it seems to ignore the - 
> mask option for untranslated use.

I shoulda dug around in the source... sorry to be a little lazy...

> It does appear that you will need to use standalone blat if you want  
> to use masking options.  Many users for untranslated don't even  
> bother with explicit masking. There is still some "masking" that  
> occurs by virtue of overused tiles being ignored and thus unable to  
> initiate a hit.

our sequences are heavy with long repetitive elements, so we're just  
trying to avoid repeat-repeat, high-scoring/match % targets without  
extra parsing, for simplicity and speed.

> It was unclear to me why you were saying that you needed gfServer  
> instead of standalone blat to run many queries.  You can have an  
> enormous set of queries with standalone blat.

For the original series of queries I was running, blat was giving me  
the out of memory/'segmentation fault' (i believe i was using a 32-bit  
binary with hg18 and/or mm9 w/ 4 GB free RAM).  gfServer appeared to  
be more friendly - thus the reason I *started* using it in place of  
blat.  Procedurally, it was also a little easier to fire off the  
queries one-at-a-time because (1) I was piping inputs and outputs  
around and I was attempting to reduce the memory load, (2) i was  
parsing the results into a database, and query by query was just  
simpler to implement, and (3) i was imagining a future scenario where  
we have different gfServers running on a remote machine(s) with lower  
RAM clients (running the code) pinging the appropriate server/port for  
their result.

I'll give blat a shot with a large set of inputs (we have groups of ~  
20,000, 200-400 mers) and see if that fits into the workflow well/ 
increases speed, which I imagine it will.  Otherwise, standalone blat  
is working perfectly (with a smaller 2bit file) for what we're doing,  
so it's looking like i'll stick with it ; ).

thanks again for your fast response and have a good one!

cheers,
-b


< * )
  (_ \\
  _ ||





_______________________________________________
Genome maillist  -  [email protected]
https://lists.soe.ucsc.edu/mailman/listinfo/genome

Reply via email to