Bayesian difference refinement

[Script for Fdiff | Keywords for Fdiff]

FDIFF is a routine to generate a pseudo-mutant dataset for difference refinement. The output dataset ("Fdiff, sigma-fdiff, weight") is written in X-PLOR format and can be read in to X-PLOR just as if it were Fobs, sigma, weight. X-PLOR automatically uses the weight as a weighting factor in refinement if it is input in this way in the structure factor file.

Fdiff is used in cases where a "WT" structure has been refined and a "mutant" dataset is available, and where it is the differences between the WT and mutant structures that is of interest. This routine takes Fcalc for the WT and mutant datasets and Fobs for the WT and mutant datasets to create a pseudo-mutant dataset: Fdiff and sigma and weights. These are then used just as if they were Fobs,mutant and sigma and weighting factors in refinement of the mutant structure. You can select reflections with F>n*sigma using RATMIN, but even if you include all reflections, a reasonable weighting factor will be generated for the weak reflections.

The value of Fdiff is given by:

Fdiff = Fc (WT) +Beta* (Fobs, mutant - Fobs, WT)

where the factor Beta is essentially the correlation coefficient between (Fo-Fc),WT and (Fo-Fc),Mutant. Bayesian difference refinement is similar to difference refinement except for the factor Beta (which is 1.0 for difference refinement).

The input file must contain Fobs, sigma-obs, and Fcalc, for BOTH the WT and "MUT" (variant) datasets. It optionally may also contain a flag marking "TEST" reflections for free-R calculations. To generate the input file, you will need to run X-PLOR or another program to get Fcalc values, then IMPORT the Fcalc values and FILEMERGE with your Fobs, sigma-obs data.

The program allows you to keep reflections flagged with RTEST>0 separate from the working set of reflections. The RTEST flag must be in a "column" of data in the input file. RTEST=1 indicates a TEST reflection, 0 indicates a reflection to use in refinement.

Basic script file for FDIFF

! -------------Script file to generate data for difference refinement -------
@solve.setup                ! standard data for this dataset
infile data.drg             ! input file with WT Fo, sig, Fc and mut Fo, sig
outfile fdiff.drg           ! output file with Fdiff, sigma,weight, RTEST
ncolfowt 1                  ! column # for WT Fobs
ncolswt  2                  ! column # for sigma of WT Fobs
ncolfc   3                  ! column # for WT Fcalc
ncolfomut 4                 ! column # for Mutant Fobs
ncolsfomut 5                ! column # for sigma of mutant Fobs
ncolrtest 6                 ! column # for RTEST indicator
fdiff                       ! setup up difference refinement
!-----------------------------------------------------------------------------

A more advanced script file for FDIFF written by Joel Berendzen( joelb@lanl.gov) that works with CNS

The file bayesdiff.ksh contains a script that does Bayesian differencing on two CNS output files and writes the output to a third CNS input file.

Keywords for FDIFF

INFILE xx input file with WT Fc, Fo, sig and MUT Fo,sig, and optional
Rtest column
OUTFILE xx output file with FDIFF,sdiff,weight,rtest
NCOLFC n column # for WT Fc (WT Fcalc) in input file
NCOLFOWT n column # for WT Fo (WT Fobs) in input file
NCOLSWT n column # for sigma of WT Fo
NCOLFOMUT n column # for MUT Fo (MUT Fobs) in input file
NCOLSMUT n column # for sigma of MUT Fo
NCOLFCMUT n column # for MUT Fc
NCOLRTEST n column # for RTEST indicator (0 if missing)
RATMIN xx minimum ratio of F/sig to read in data (default=0)