Modeller will take a protein structure as a starting point, consider as a template, and that will be most likely your wildtype protein. Then, based on the mutated sequence of the protein you will provide, it will generate a 3D structure where all mutated residues will be replaced. Side-chains of the new residues will be optimised to comply with their surrounding environment but very few conformational changes will be introduced. There will be mainly local changes and no large scale conformational changes are to be expected.
So I would only consider this output as a very good input for an MD simulation. This can maybe give you some basic insights on several local arrangments of the protein for instance but nothing more. I’d advise to always couple that with a MD or a docking for instance if you’re studying the effet of the mutation on a ligand binding.
Best regards,
Mikael