More efficient blat algorithm

[blakeln]

Hello- I am using blat to match a set of sequences against the human genome. The problem is, for my particular research, I don't care WHERE or HOW MANY TIMES the individual sequences match with the human genome (which is information I can receive from the output .psl). All I care about is IF a particular sequence matches anywhere (even just once).

If it does match at least once, I will deem that sequence "human" and will then separate the non-human and the human sequences (non-matched vs. matched).

Since I am inputting close to a million sequences, it would probably save hours of computational time to change the code of blat to STOP searching for where ever else a sequence matches up after it finds just one match, then output just that first match to the output .psl file, and move on to the next sequence.

I've been searching the blat website to see if this capability exists and have found nothing. If I am confident it does not exist, then I will try to change the source code of blat to accommodate my needs, but I first wanted to see if anyone on this forum has heard of this being done already before I spend time on it.

Please respond if you know more about this.

Thanks.

[Richard Finney]

I ~think~ you can hack the gfFastFindDnaHits() and gfStraightFindNearHits() functions in genoFind.c

you can "break" on first hit ("hitCount >= 1").

[Brian Bushnell]

Are you trying to isolate human reads, or trying to remove human contamination, or just doing quantification? I developed a protocol for removing human contamination, which is very fast, for short reads:

http://seqanswers.com/forums/showthread.php?t=42552

As described, though, it only works for sequences of at most 500bp.