I am trying to do ab initio prediction of a 62 aa protein. The models generated without any restraints do not seem to converge towards a particular structure (which is understandable). This protein has an indication of being structurally similar to another protein (36 aa long) for which I have already done the modeling and the model seems good. Since the sequence identity between them is very poor (~12%). I am skeptical in using this model for homology modeling but I would like to use this model to get structures similar to the already generated model.
So the question is that - Is it possible to use structure model (not in database) while fragment generation along with the usual database and use this fragment library in ab initio prediction?
I would encourage you to generate constraints based on the model you like, and apply those constraints during ab initio. Unless you are confident that the fragments you need are not present in your fragment set, adding a few extras seems like an indirect and inefficient way to influence the result.
To answer two related questions: yes, it is possible to generate fragments from your structure (but I don't know how). No, I am not aware of an automated tool for doing this constraint generation except those already part of the homology modeling pipelines.
I think you should also just go ahead and do the homology model using your low-identity template as a template, unless computer time is much more expensive than your time. (In other words you can probably just fire that off quickly to see if it works before fiddling with a more complex protocol).
Thank you so much for the detailed reply.