[Bioperl-l] Gap Penalties for BLOSUM62

[William R.Pearson]

The most effective gap penalties for searching with BLOSUM62
were shown in:  Pearson (1995) "Comparison of Methods for
searching protein sequence databases" Prot. Sci. 4:1145-1160.
In that paper, I showed that the best gap penalty for BLOSUM50
was -10, -2 (reported as -12, -2, because the first parameter
was not a gap-open penalty, but a first-gapped-residue penalty)
and the best penalty for BLOSUM62 was -7/-1 (reported as -8,-1).

The -7/-1 gap penalties for BLOSUM62 are very similar to the
-12/-2 used for BLOSUM50 (by most programs), because BLOSUM62
is scaled in a way that a +6 or -6 in BLOSUM62 is equivalent
to +9/-9 in BLOSUM50.  Hence a different scaling for the gap
penalties.

The BLAST and pSW penalties reduce the sensitivity of the
BLOSUM62 matrix; I suspect that BLAST uses -10/-1 because of
the way BLAST statistics are calculated; -10/-1 will be much
less likely to ooze towards global alignments for unrelated
sequences. I think Ewan would have a difficult time defending
-12/-2.

Bill Pearson