I normally prefer to do the modeling myself, using a variety of different programs, rather than automated servers. The methods used depend on the question I wish to address with my model and on the level of difficulty the particular model presents. I find that a very hands-on approach to modeling helps me to discern details that I might overlook when just examining the final model. Depending on the difficulty of a project, I might try different methods and compare the results. Template choice might not just depend on the highest similarity/best resolution, but also include factors like: does this potential template agree with the consensus of related templates, or does it show obvious deviations. If so, can these be explained, e.g. by sequence differences, ligands present, crystal contacts, crystallization conditions.

50-60% similarity normally is OK, although the devil is in the detail. It normally means that you have a quite highly conserved protein core, while loops and surface residues are much more variable. I like to color a 3D alignment of potential templates by similarity to the target sequence to examine where sequence deviation are located, and whether they are likely to have any effect on the conformation. This also helps to verify whether the placement of insertions/deletions make sense.

I prefer to do much of my analysis before any energy minimization, since minimization tends to blur problems by spreading the bad energy over a larger number of residues.

One of the problems with membrane protein structure prediction is that most automated modeling programs are targeted at and calibrated to small , soluble, globular proteins. Their algorithms favor exposure of charged and hydrophilic residues, and burial of hydrophobic residues. So even if template homology leads to a satisfactory overall structure, quality check statistics look bad due to the large hydrophobic surface area in membrane proteins.

There are some sites that explicitly target membrane proteins which you might try, e.g. the tools for membrane protein prediction at:

http://opig.stats.ox.ac.uk/webapps/medeller/index.html, part of Memoir:

http://opig.stats.ox.ac.uk/webapps/memoir/php/index.php

For a more "hands-on" approach rather than just submitting a sequence to a server, check out ROSETTA Membrane homology modeling:

https://www.rosettacommons.org/demos/latest/public/membrane_homology_modeling/README

For a publication on the challenges of membrane protein modeling and some useful links to "membrane-savy" modeling tools, see Article Computational modeling of membrane proteins

(To read the article, go to https://www.ncbi.nlm.nih.gov/pmc/articles/ adding PMC4270820 to the URL, since Research Gate captures the URL despite not having a link to the full text)

Did you do homology modeling? Was your alignment refined or you took what the software gave you? Did you use a forcefield on your model?

thank you Dr. Annemarie for your elaborated answer. Given your experience in membrane protein modelling what do you personally prefer Rosetta or modeller server?

Hey Mr. Jorgaq.

I have used itasser output for homology modelling. I didnt refined the alignment. i used the given structure in gromac or charmm forcefields for simulations.

I have an additional question is homology of 50 - 60% fine for homology modelling? or i should use ab initio methods in these cases?

I normally prefer to do the modeling myself, using a variety of different programs, rather than automated servers. The methods used depend on the question I wish to address with my model and on the level of difficulty the particular model presents. I find that a very hands-on approach to modeling helps me to discern details that I might overlook when just examining the final model. Depending on the difficulty of a project, I might try different methods and compare the results. Template choice might not just depend on the highest similarity/best resolution, but also include factors like: does this potential template agree with the consensus of related templates, or does it show obvious deviations. If so, can these be explained, e.g. by sequence differences, ligands present, crystal contacts, crystallization conditions.

50-60% similarity normally is OK, although the devil is in the detail. It normally means that you have a quite highly conserved protein core, while loops and surface residues are much more variable. I like to color a 3D alignment of potential templates by similarity to the target sequence to examine where sequence deviation are located, and whether they are likely to have any effect on the conformation. This also helps to verify whether the placement of insertions/deletions make sense.

I prefer to do much of my analysis before any energy minimization, since minimization tends to blur problems by spreading the bad energy over a larger number of residues.

Yogesh Sharma 50-60% of identity is good, the alignment musn't be so bad, but it is always good to check it yourself, especially in the case of membrane proteins. You should take care to the TM helices, to align the beginning of them, otherwise it can cause you some problems. Check the loops too, and after that you can go more into details with your model.

I would prefer to go with I-Tasser and do a 20-30ns simulation in water to remove any steric hindrences in the structure.

If pore region and NPA motifs are perfect, I-tasser will work for Aquaporins.

Nikhil Maroli

Sir

what if homology modelling in itasser modelled it to available closed structure. I wont be able to open it i guess.?

Yogesh Sharma Did you tried to model with I-Tasser? your resultant protein is in closed shape? which aquaporin you want to model?

Singh RK, Shweta S, Muthamilarasan M, Rani R and Prasad M (2019) Study on aquaporins of Setaria italica suggests the involvement of SiPIP3;1 and SiSIP1;1 in abiotic stress response. Functional & Integrative Genomics 19: 587-596.

Yogesh Sharma please send me the PDB file.

Homology modeling for the covered regions; the rest depends on the distribution of the covered regions: homology modeling for the non-covered regions if they are short or other approaches if the non-covered regions are too long.

Martin Klvana

Do i have to model protein in fragments? for non covered regions. if it so? how will I merge it with core?

Yogesh Sharma , build a continuous homology model of the entire protein (excluding the uncovered termini), and then, if needed, remodel (using other approaches) the obviously badly modeled regions -- if the badly modeled regions are solvent-exposed, not buried in the core, it should be doable.

Martin Klvana sir, using modeller i could model transmembrane part with high confidence but loop reasons are still poorly modelled. I tried sphinx loop refining for loop regions but of no use. can simulation solve the rest of structure?

and even i am not able to upload generated structure to charmm gui. getting error "missing residues" i found out structure doesnt have hydrogens. what can i do here?