Pavel Afonine wrote:
Hi Ed,
sorry for the trouble. The error you get is a memory problem indeed.
"fix_rotamers" is not a strategy. If you do:
phenix.refine --show-defaults=all
it will print out all phenix.refine parameters and you will see the list of "strategy" keywords:
strategy = *individual_sites individual_sites_real_space rigid_body \ *individual_adp group_adp tls *occupancies group_anomalous
where "*" means selected.
"fix_rotamers" option may consume a lot of memory since it is using 3 maps: mFo-DFc, 2mFo-DFc and Fc maps. All these maps are computed on the grid with the step high_resolution/4. I found that the scale 1./3 sometimes works worse than 1./4, although of course using 1./3 will result in smaller maps.
Thanks, Pavel, If nothing else works, can I change this factor with: real_space_refinement { mode = simple *diff_map target_map_name = 2mFo-DFc model_map_name = Fc grid_resolution_factor = 1./4 ? Or its hard-wired for all three maps?
Currently you can't select residues to apply "fix_rotamers". phenix.refine automatically decides which residues need a fix. The details of "fix_rotamers" protocol you can find here:
http://cci.lbl.gov/~afonine/rsr.pdf
My plan is :
- find a way of using less memory since it seems to be an issue for small memory machines/large structures; - enable user-defined atom selection so one can apply "fix_rotamers" to selected residues only; - improve quality assessment criteria to reduce false fixes.
I was working on this option beginning of this year and then I got some sidetracks so I still consider it "under development". For example, it doesn't play very well if you use NCS restraints. At low resolution I need to experiment with B-factor sharpened maps or/and 3Fo-2Fc or higher order, etc. More efficiently use H atoms.
Anyway, if you try "fix_rotamers" option I would be interested to hear you feedback. Thanks!
All the best! Pavel.