Finally got some decent models out of the computer.
My problem is processing time. Because of the way I build up my protein models, each amino acid residue adds a factor of 18 to the number of conformations. That is, I have 18^N conformations. Somehow these have to be searched to find the minimum energy structure.
However a molecule doesn't exist in only one conformation, except at zero Kelvin or when crystallised. It moves about in the solvent. We hope to capture this. What we do is run it in a simualted temperature bath, until it has found the minimum conformation. Then we run it for as long again, taking a thousand samples from the run. We then super-impose these to gain a view of the conformation.
This leads to the question, how do we know it has found the minimum? What we do is run three independent simulations. if all three find the same minimum value, we assume that the genuine minimum has been found.
Here are the results for a bovine rhdopsin segment (protein data bank code 1eds).
The first image shows the native structure. The second image shows the results using the forcefield as it used to be. The final image shows my "improved" forcefield which includes rotamers. Rotamers give the sidechains of the amino acids a bit of freedom to wiggle about. The old forcefield treated them as a rigid stick of balls.
What has happened is that the rotamers have increased the entropic space of the problem enormously. We go from 6^N moving just the protein backbone to 18^N moving sidechains as well. So for the same temperature, the simulation is fuzzier.
Eyeballing it, there seems to be no real difference in accuracy. We've more or less got the alpha helix at one terminus, but the less structured part at the other terminus is not modelled very accurately.
The image you see took about three days with 8 processors. The problem is that I had to run three replicates to make sure I'd got the minimum, then the old forcefield as a control. So in fact it tied up all my processor allotment for 3 days. So I've got a new account on a bigger machine. It has a job limit of 48 hours. However I've bumped up the number of processors to compensate.