In general 65% identity is good enough for modeling.

But you must look for the coverage and alignment of the target and template seq.

However, the usefulness of the template should be assessed by the evaluation of the calculated 3D model. This is the best approach, because the evaluation of a 3D model is generally more sensitive and robust than the evaluation of an alignment.

It depends on the structure of the target or template. You have to look through the alignment strings to estimate if these 65% of identical residues are evenly distributed on all length of the sequense. If it doesn't, the areas with poorly similar residues ccould be additionally aligned against the rcsb database (with blast for example) and rebuilt from the another template. It seems, that the lowest threshold for the homology modeling is 35-40%, thus Your 65% value looks nice.

I generated the models using modeller and then check the models with PROCHECK. All of them show above 95% in allowable regions in ramachandran plot. Is that acceptable.

Hi Pratik

Try MolProbity also for structure validation..

Apart from identity look at the the query coverage also, as Alex has already mentioned. If the query coverage is good enough and most of your protein's sequence is covered, especially the functional domain, then 65% is good enough with adequate model refinement. In case the 65% homologous structure doesn't cover your full query protein length, and some specific domains are missed, you need to go for multi-template modelling using more than one template structures which together cover all the relevant domains of your query protein. It's fairly simple and you'll find plenty of tutorials online. Remember in either case, you'll need to refine your model extensively, by energy minimization, loop refinement, and ideally MD simulations.

In general 65% identity is good enough for modeling.

But you must look for the coverage and alignment of the target and template seq.

However, the usefulness of the template should be assessed by the evaluation of the calculated 3D model. This is the best approach, because the evaluation of a 3D model is generally more sensitive and robust than the evaluation of an alignment.

Hi Pratik,

If you have good query coverage (as close to 100% as possible) then I would say that 65% is definitely good enough for modelling. Have you tried other homology modeling techniques to validate/compare the strucuture? SwissModel is another freely available resource that you can use. I would also look at the SCOP fold of your model. Is this a highly conserved structure? If your target of interest is a highly characterized protein (e.g. a kinase) then you should have a good idea of what the 3D structure should look like. Post modeling analysis can be performed using Ramachandran Plots. You can manually minimize any areas that are 'disallowed' also. Please refer to this publication also. An interesting read for this area: (Burkhard Rost, “Twilight zone of protein sequence alignments”, Protein Engineering 12: 85–94, 1999)

Julie Owen : Thanks.

yes, it is a highly conserved structure. The proteins are form GLUT family. I have GLUT1 from PDB(4pyp). I am trying to model GLUT2, GLUT3 and GLUT4.

I am not sure how to do advanced modeling. I performed modeling using simple modeling tutorial given on the website.

Regards,

Pratik

The homology model I generated has a region (last loop) which should be a helix but is a simple long chain. How can I covert it to a helix?

Pratik, try using other modeling resources and compare results. These may accurately predict the helical structure. If not, then you can try modeling the exact sequence you think is the helix? But that will involve a lot more post modeling work. I use MOE or ICM Molsoft for that. There’s a free browser for ICM Molsoft and lots of webinars available freely online. Two other resources you can use are: https://zhanglab.ccmb.med.umich.edu/I-TASSER/ (frequently does well in CASP) And https://swissmodel.expasy.org (very easy to use).

Thank you.