How to predict a 3D structure more precisely?

Currently, I've been working to a transmembrane protein that contains an extracellular domain and an intercellular domain. However, I am only interested in the extracellular domain.

Therefore, is it better to predict the whole protein and then export the extracellular region from it or only predict the extracellular domain directly?

PS: some parts of the extracellular domain have crystal templates, while there is no experimental information about the intracellular domain.

Thanks a lot.

Ivan