STARTING and ENDING model variables define how many models to generate. Because the structure optimization is stochastic (MD simulation) you can expect minor differences among different structures, especially if you have such regions in the structure which are not very well defined by restraints (segments with insertions, vicinity of deletions or loosely defined substrates relative to the template etc.).
These variables also define the extension of your model files as it was noted by Azat: technically it is useful if you run modeling under different circumstances, e.g. generate 10 models with one ligand between 1-10, than generate 10 models using different ligand or templates (Starting MODEL=11, ENDING_MODEL=20) etc.
> I am modeling a very short peptide with modeller4. If I set > STARTING_MODEL=1, ENDING_MODEL=10, I get 10 models which have different > orientations and spatial location. If I would like to get a model which > can overlap with the original model (maximum uses the template > coordinates), how to do it? And is it possible to keep the backbone in > the same conformation of original peptide ?
if you include in your top file: SET FINAL_MALIGN3D = 1
modeller will automatically superpose all your models onto the template and generate the *fit files which contains strcutures whith supeprosed coordinates.
or you can write a short top script to superpose any number of structures from any sources generated, as explained in the manual under MALIGN3D and SUPERPOSE commands.