Dear all, I know this has long been a tricky thing to do, and in the past I've always managed to model multiple alternative conformers by hook or by crook, but in some current structures I'm having a bit of a nightmare getting things to refine with reasonable geometry, and was wondering if anyone has noticed this and/or managed to resolve it satisfactorily.
If you add an alternate conformer in coot, you cannot then do real space refinement with either sphere refine or real space refine zone (eg i +/- 1 by pressing 'a' or by selecting a zone around the residue with multiple conformers). In the case of sphere refine, the i-1 backbone peptide will become non-planar (or completely flip so the C=O bond overlaps the C-N bond), and the i peptide will also be distorted though to a lesser extent. In the case of real space refine zone, the i-1 and i backbone angles are distorted such that C-N-CA is around 150 degrees. You can real space refine zone, then just select one of the multiple conformers, which has worked acceptably in the past (especially when then fed into a refinement package, which tends to clear things up), but at least on the several datasets I've tested it on, this is still leading to distorted geometry, particularly peptide planarity of the i-1 and i peptides. I'd really like to get a model with good geometry directly from coot, so I don't have to rely on maximum likelihood refinement to sort out the problems - I always find that a rather clunky solution. Has anyone else noted or documented this behaviour, and is it possible to fix this? I know I may be asking a lot, here! To clarify some of the obvious questions up front: I'm using coot 0.8.1 on Ubuntu 12.04 and coot 0.8 on Max OSX with the same results. The observed behaviour occurs with and without the following restraints in coot: planar peptides, torsion and Ramachandran restraints, with a refinement weight typically of about 10, but I've observed the same problem at any value between 0.1 and 90 (I haven't tested wider as there's not much point!). You don't need high res data with genuine alt conf density to test this, you get the same behaviour by modelling in 2 x 0.5 occ conformers into the same density. The only work around so far seems to be real-space refining without the second conformer to get all the geometries right, then putting in the alt conf and selecting the nearest rotamer, then do ML refinement to try to fix the errors and get B-factors. Deleting the alt conf in coot before any additional real space refinement is then required (but coot cleverly remembers the B factors for the alt conf when you add it back in, which is neat). I'd be extremely grateful for any help to resolve this! Thanks in advance, and thanks to the coot team for their hard work making this indispensable programme! Isaac
