Sendu Bala <bix <at> sendu.me.uk>
2006-08-02 13:03:36 GMT
Hilmar Lapp wrote:
> On Aug 2, 2006, at 7:23 AM, Sendu Bala wrote:
>> It should be noted that all frac_* statistics and probably others from
>> hit objects have had a high chance of being wrong in the past, and
>> frac_identical and frac_conserved can still be very wrong*. It would be
>> a good idea if someone were to make these methods return slightly more
>> sane numbers.
>> [*] which is to say, more wrong than you might reasonably expect, given
>> the limitations with gapped or alignment-free blasts
> Can you elaborate? Specifically, can you put in tests that show the
> wrong results?
I've added some new tests based on Andreo's blast result, but atm I've
left the tests commented out. See line 882 of t/SearchIO.t revision 1.94
- a sane result would be less than 1.
I think it ought to be possible to get better answers, but I got the
feeling the fix wouldn't be completely trivial so I let it go, not
having the time to spare right now.
The reason I don't just call this a bug and make a bug report is that
the documentation acknowledges that you won't always get a good answer,
so it needs to be investigated if the current answer really is the best
that can reasonably be given, or if there is some bug making the answer
worse than it needs to be (as was the case with frac_aligned_hit and