Hai,

You can try modelling your protein using multiple templates. Already 80% of your protein has a crystal structure with 100% identity. Blast the remaining 20% and find a better template. Use both the templates to model your protein. For modelling you can use software like modeller or Prime,Schrodinger (Which has detailed tutorial to perform multi-template modelling).

Thank you and best of luck.

I would advise performing a secondary structure prediction on the whole protein. Once you see what type of secondary structure the missing parts are expected to have, you can model them in with Modeller. As Lakshmanan suggested, using multiple templates is also a good idea, both in the modelling and secondary structure prediction stage (i.e., see what secondary structure similar proteins have).

Have a look at these:

http://www.compbio.dundee.ac.uk/jpred/

https://salilab.org/modeller/tutorial/basic.html

Good luck.

Depends on whether the residues that are lacking in your template can be extracted from another template. Look up for pGenThreader and similar threading web servers to find out other possible templates. If there is no structure information available, then you would either have to disregard that portion of the protein in your final model or add it using comparative modeling in combination with ab-initio/statistical modeling (eg. Rosetta)

I would recommend you to cut this part of sequence and isolate it, then you can do a search to find similar proteins (whose 3D structure is known) and use any homology modeling software (I strongly recommend you Modeller), once you have modeled this part of your protein (and checked out if it is correct by assessing Ramachandran plots...) you can join it to the rest of protein you have. Finally perform a minimization (or mollecular dynamics symulation) and evaluate if your model have good statistics.