Thanks to everyone who replied.
Doing this with HL coefficients just prior to solvent-flattening was
the least painful. I arbitrarily divided by two (see below).
Most importantly, the maps improved significantly, showing clear
nucleotide density for missing (unmodeled) RNA fragments where before
there were only uninterpretable blips.
On Jul 11, 2008, at 11:47 PM, Randy J. Read wrote:
Hi Bill,
The easiest thing is to scale down the HL coefficients, e.g. by
dividing them by two. (Dividing by two has the effect of taking the
square root of every value in the phase probability curve then
renormalizing, which reduces the sharpness of the probability
distribution without changing the positions of the peaks. It's also
equivalent to increasing the underlying variance of sources of error
in the phasing.)
You could do this in sftools. It's likely that in some programs
there's an option to provide a scale factor for the HL coefficients.
Regards,
Randy
AND ...
On Jul 11, 2008, at 11:36 PM, Thomas Edwards wrote:
this sounds exactly what "blur" hl coeffs does in CNS.
From their web pages:
hlcoeff_blur.inp
"Blur" Hendrickson-Lattman coefficients for use in refinement with
the MLHL target
The phase probability distribution is "blurred"
by application of a scale factor (S) and a B-factor (B):
HLA_new = S * e^(-B*s*s) * HLA_old
HLB_new = S * e^(-B*s*s) * HLB_old
HLC_new = S * e^(-B*s*s) * HLC_old
HLD_new = S * e^(-B*s*s) * HLD_old
This is performed to compensate for overestimation of phase
accuracy which most often occurs after density modification
or when probability distributions are derived from an
atomic model.
Warning: this should only be used when there is a good
reason to believe the phases are biased. MAD or
MIR
phase probability distributions should usually
not be modified in this way.
Good Luck!
Cheers
Ed
