On 08/04/2009 12:42 AM, Robert Jorissen wrote:
I am modelling a complex of two proteins, each of which has moderately high sequence similarity to their respective template proteins. I wish to create a model that strongly resembles their targets. The idea being that I wish to demonstrate the plausibility of the interface between the two template proteins being conserved in the target proteins. What is the best way to go about this? Presently, I have created a number of models using the library_schedule autosched.fastest with a starting coordinate deviation of zero (i.e. no random displacement added to the initial set of coordinates) and taken the best scoring of these. Is there a better way that optimizes the MODELLER score, but that does not stray far from the template structure, particularly in the ungapped, high similarity regions?
Modeller aims to generate models that stay close to the template in aligned regions, by construction (since the restraints are derived from the templates). So you shouldn't have to do anything special. If you want to minimize the chance that the optimizer gets confused and wanders away from the template, what you've done sounds reasonable. However, if you set deviation to zero, so there's no randomization at all in the starting conformations, there's no point in building more than one model per alignment - since the modeling process is deterministic you should get exactly the same model (and the same score) each time.
Ben Webb, Modeller Caretaker -- http://www.salilab.org/modeller/ Modeller mail list: http://salilab.org/mailman/listinfo/modeller_usage