Most likely, your protein consists of multiple independent folding units (domains), possibly interspersed and connected by flexible regions, that may even comprise stretches that are intrinsically unfolded. Check the UniProt entry for your protein and its homologs in the closest model species for what is already known about its domain organisation - it may even give you a link to an alpha-fold model. If not, you may try to generate a global model using alpha-fold, and compare this result to the models you get with other methods. Caution: You need to pay attention to the confidence information, as regions that have no defined structure in real life will be assigned coordinates that show as loose loops with low confidence.

Frequently, when experimentally determining protein structures, you have to produce constructs that only comprise the folded domain. Check out the publication for the available experimental structure - The authors may have discussed why they solved just this part. Most likely, this is a very good template for this part of your structure.

As automated homology modelling and threading programs normally do not work with multiple templates, you may want to submit only the sequence not covered by the initial model to try to find templates for these parts.

When considering the suitability of a template for protein modeling, it is important to assess both the query coverage and similarity. While a high similarity (100% in your case) indicates a strong sequence similarity between your protein and the template, the low query coverage (24%) suggests that only a portion of your protein aligns with the template.

In such cases, it is worth exploring alternative modeling techniques to obtain a more accurate structure prediction. Here are a few options you can consider:

As for relevant servers or software, here are a few commonly used tools for protein structure prediction:

• SWISS-MODEL (https://swissmodel.expasy.org/)
• Phyre2 (http://www.sbg.bio.ic.ac.uk/phyre2/html/page.cgi?id=index)
• I-TASSER (https://zhanglab.ccmb.med.umich.edu/I-TASSER/)
• ROSETTA (https://www.rosettacommons.org/software)
• Modeller (https://salilab.org/modeller/)

These tools often provide a range of modeling approaches and can help you assess the suitability of different methods based on the available data.

Ultimately, the choice of modeling approach depends on the specific characteristics of your protein and the available resources. It is recommended to assess multiple methods and compare the results to gain confidence in the predicted structure.

Quite honestly if your protein isn't too large, i.e., to many amino acids for it I would just use AlphaFold or ESMFold and compare the best model with the resolved one by aligning on this region. I think the models (or variants of it that participated listed in the previous post all do have lower performance in the last CASP competitions then AF had. Although I haven't checked this ^^

RosettaFold would also be a good option.

Of course homology modeling can still work pretty well, but usually only if you have good templates. But if you have regions that basically are missing in your templates and those are significant it usually doesn't really work that well.