*** For details on how to be removed from this list visit the *** *** CCP4 home page http://www.ccp4.ac.uk ***
James: Doesn't this ignore the use of B-factor restraints in the refinement programs? Because the B-factors are in most cases refined in this way, modelling an atom in a position not supported by experimental evidence, and allowing its B-factor to inflate, will artificially inflate others in the residue as well, and correspondingly, will restrain its B-factor to a value lower than it should be. Well, yes! B-values are restrained for a reason, being that the atoms have covalent bonds! The crystallographers choice of (lowering) the weight on this restraint is in the end an interpretation of the data, just like leaving out atoms that are really there. Models are always wrong of course, the debate is what is more wrong and what is less wrong... James: Furthermore, this practice would be expected to greatly increase user error because there is no clear indication of the atoms that were positioned arbitrarily according to non-experimental terms. Hopefully the end-user will look at the model in the context of the structure factor data, but as we know this is not always done, and even worse, can't be done where the structure factors haven't been deposited. In this worse case scenario, how can someone be sure that any surface interactions between sidechains with middle-to-high B-factor is supported at all by the electron density? Hopefully the crystallographer who models missing atoms in this way is vigilant about depositing their structure factors! A measured reflection contains contributions of all atoms in the structure, including the bulk solvent. The bulk solvent molecules are also not visualized in the final model, but are taken into account during refinement anyway. If you leave atoms out of the refinement, you introduce or at least spread out errors on the remaining atoms. Not depositing structure factors is definitely a nono, I agree. Maybe we should also publish all the restraints etc in the remark section as well, but that certainly won't be read by the majority of users! James: I think building atoms in places not supported by the data without at least an occ=0 is the wrong approach. There are plenty of modelling programs out there that can do that for you, even humble SwissPDB viewer will arbitrarily reconstruct missing side chains and colour them differently to the rest of the protein. The atoms are placed based on experimental evidence (neighboring atoms), plus geometry restraints (including rotamer statistics). Again, model error can only be assesed for the model as a whole, not just for individual atoms I think. Flip -- Dr. James Irving NH&MRC C.J. Martin Fellow Division of Structural Biology Wellcome Trust Centre for Human Genetics Oxford University Roosevelt Drive, Oxford OX3 7BN UK email: [EMAIL PROTECTED] phone: +44 1865 287 550
