On Sat, Nov 19, 2011 at 1:20 AM, Tim Fenn
However, a purely repulsive potential should be able to yield good results! Perhaps the original poster can try a different weighting scheme?
We still haven't heard details from the original poster about the version used, etc., so it's somewhat premature to draw conclusions about the causes and/or remedies. Previous versions of Phenix used a weight on the repulsive terms that was at least 6x too weak, with the result that clashscores (and Ramachandran scores) were often substandard at low resolution. This has been fixed, and with the current version it is very rare to see a clashscore above 60 at the end of refinement *unless* the starting model is severely strained, and models with good packing are rarely made worse. With proper rebuilding and weighting it is entirely possible to refine a sub-4 Angstrom model with Phenix to a clashscore below 20 (the Molprobity-recommended maximum) without any additional restraints or tricks. However, in all of our tests there is a limit to how much improvement one can expect from minimization alone - and even really aggressive sampling like what Rosetta does has its limits. The geometry restraints aren't the limitation here (Rosetta uses all atoms and attractive forces too), the optimization techniques are. -Nat