[Genome] blat: fastMap, semi-global vs. local

[Galt Barber]

BLAT, like BLAST, is probably better defined as a Heuristic aligner.
We think of them as local alignments.  While BLAST only aligns exons,
BLAT finds exons from the hits on diagonals, and then also
has a post-processing step that chains the exons together
into a complete alignment.  For DNA (not protein) at least,
the intron boundaries are explored for improvement by slight shifting.

When -fastMap is specified, the chaining step will not take place.
Thus no multi-exon alignments will result.  You can also make
BLAT produce single-exon alignments if you specify that the output
be in BLAST format.  I have found that using -fastMap with very short exact
sequences can cause it to lose sensitivity, perhaps due to a bug,
so I recommend not using -fastMap for those.

Once an alignment candidate hit is found, BLAT runs more traditional
dynamic programming algorithms with some heuristics to extend the
alignment from the seed hit(s) to the full "exon".

Please read the full BLAT FAQ.
There's lots of useful information there.
 http://genome.ucsc.edu/FAQ/FAQblat

-Galt


On Wed, 23 Jan 2008, Isaac Ho wrote:

> Hi there--
>
> In default mode, it appears blat attempts to make a semi-global
> alignment of the query.  But in fastMap mode, I've noticed lots of local
> alignments.   Is this indeed the case?    Can you shed more light on
> what fastMap entails--specifically how mismatches and gaps are
> handled?    If fastMap is on (no introns are allowed), and only perfect
> matches count for tile hits, and maxGap = 0, then I assume that only
> tiles on the same diagonal can be stitched together ( meaning
> effectively there must be either no or cancelling-out indels  )?
>
> Thanks,
>
> Isaac
> _______________________________________________
> Genome maillist  -  Genome at soe.ucsc.edu
> http://www.soe.ucsc.edu/mailman/listinfo/genome
>