PWM search: one more bug is now fixed
COTRASIF
("... the solution was to apply a minimal value of 10^-9 to all
zero-value cells in the submitted fractional PFM... ") introduced a
problem, which became apparent while testing the fix for the very
recent fractional PFMs handling bug.
The problem manifested itself in effective cut-off being much lower
than requested and displayed in the results file. This was caused by
the "stretched" range of minimal and maximal PWM model values,
stemming from the use of near-zero pseudocounts for fractional PFMs.
This bug was unknowingly introduced on 2009-06-19.
It affected 10 tasks (by 8 submitters), including those 3 tasks by 2
submitters mentioned in the previous bugfix report.
Affected task results are still valid, but their indicated binding
site similarity scores are wrong (much higher than real).
These conditions had to be met for the bug to manifest itself:
- one had to use a fractional PFM
- that PFM must have had exact 0.0 values in it
- the task had to be submitted between 2009-06-19 and 2010-04-14
Current fix is not exactly perfect, but very close to that.
Integer/count PFMs (which represent the majority of submitted tasks)
were *not* affected by these two recent bugs.
Sorry for any inconveniences.
Bogdan