I am generating models for a strained protein chimera using weakened
restraints on one of the two domains, to see if Modeller will unfold the
domain to relax the mechanical strain. I have weakened all Ca-Ca and N-O
distance restraints (roughly 3,800 out of 16,500 total restraints) by a
factor of 100, but it had nearly no effect on my output models. Is there
something fundamentally flawed with this approach? Are there other
specific restraints that might be recommended for weakening, or would it
be best to simply weaken all restraints associated with the targeted
domain?
This may just be a sampling issue. Remember that the starting structure
for comparative modeling is the target sequence threaded onto the
template structure, so if you want to deviate from the template you may
have to overcome local minima first. Longer simulations and/or different
starting conditions may help you here.