Energy Minimisation of Proteins in vacuo

 Hi CCL'ers,
 I have been trying for a while to refine a protein structure (from a PDB !)
 and try and remove the problems of poor dihedrals (not just due to
 function), poor bond lengths, iffy sidechain planarity etc so it can be
 used to do rational design, de novo design etc etc.
 Now the protein I have to work on is 70Kda or so and to try and do this
 work using a solvated model would be fairly mad because I would have to buy
 a few extra hard drives for my indy to process it with a newton raphson
 type algorithm.  The alternative in literature suggests in vacuo
 simulations using energy minimisation (DISCOVER, CHARMM, MAXIMIN whatever)
 with a variable dielectric-distance function and with a fiddle factor
 dielectric (different authors reckon either 4 or 20). I tried all these
 with my protein and it seems to work a lot better at 20 rather than 1 - and
 according to the Vriend & Sander WHAT_CHECK for proteins the dihedrals
 (according to ramachandran) seem to have become more tightly focussed where
 they should be and the side chains more planar, bond lengths seem to be
 more like the norm.
 Is this a good alternative to solvation or does this technique completely
 over-refine the model and make it worse that just using the raw data ?
 Any comment would be greatly appreciated, summary as usual.
 Best Wishes,
 Richard Day.
 Cranfield Biotechnology Centre,
 +44 1234 750111 x3562