I'm working on nsSNPs(Non-synonymous single nucleotide polymorphisms) of protein AKAP13 whose PDB not available. I did one mutation in this protein and I want to show the effect by 3d modeling without using a PDB but don't know how to do. Can anyone help Please!!
Getting a crystal structure is hard work for crystallographers, even with a high throughput systems, so they make constructs with only known domains or regions of particular interest. What happened here is that only two parts have been cloned, expressed and crystallised. The rest may be crystallisable or solvable by cryoEM, but may be too disordered.
If the protein has a close homologue it can be threaded, that is it's sequences is "threaded" over the structure of the homologue. Two site provide these ready-made.
The former is very clear, has an NGL viewer, a handy circular map and the assessment is clear. For your protein it is showing four existing crystal structures, but these cover only a small part of it. The homology modelled section is small and not good here, but handy as the model has two numbers Qmean of –4 and identity of 23%, two red flags. So the next place to check are the following servers. They thread against different structures and hybridised. Missing parts are predicted ab initio (from the scratch). The latter parts look like spaghetti and cannot be trusted.
- Phyre2. <1 day turnover. Do pay attention to the colour rails. If your region of interest in low confidence and is based on nothing, then do not trust it at all. Also there it is good to double check keep an eye on the sequence alignments of your position as misalignments can happen.
- I-Tasser. 3-5 day turnover. The CAPS server winner year after year. You 5 models representing each cluster. If they look very different and the metric is bad for the top, then you have very low confidence and must look at the all if even.
- EVFold. The amino acid co-ocurrence is rather interesting
- AlphaFold... Google keeps it secret. Nobody except them can use it.
Here is an example of a spaghetti model (not your protein):
The except for the red bar, which is modelled against homologues and forms a nice domain within the spaghetti ball at the bottom, all of the rest is ab initio and should not be trusted as the warning in the "Multi-template and ab initio information" section says.
A nice addendum to a model is Consurf, which maps conservation to the models as B-factors. In a crystal structure, the core is conserved and the surface is not. Metrics are nice, but human insight is better.
A big problem are protein that are part of a complex, because alone they are not structured and their structure is dependant on a neighbour. Models are great if they are threaded against a model that forms the same complex, eg. mouse to human, but not if they are hybridised.
In the case of multidomain protein that are large. It is best to split the protein up. Furtehrmore, I-Tasser and Phyre have length limits because the longer the protein to predict is, the higher the chance of error. One can determine where the domain boundaries may be —a Blast or a few iterations of Psi-blast will show conservation boundaries nicely. PFam has a lot to say about the protein architecture: enter link description here Regions of high disorder cannot be and should not be modelled as they do not form a structure free in solution —unless wrapped around another protein as mentioned.
Wisdom to step away
However, the real question is not
how to but
should one. If one only get ab initio results, do not trust them at all —count your losses as opposed to embarking on a castle in the sky. For a SNV, check if it is close to a post-translational modification in PhosphoSitePlus or if there is a linear motif in ELM.
Answer from @o-laprevote, converted from comment:
Your protein includes intrinsically disordered regions (https://www.uniprot.org/uniprot/Q12802#family_and_domains). Showing a predicted structure may give a wrong idea of what it looks like: I'd consider simply showing the ordered part (of which a structure already exists) plus a "spaghetti" starting from its N-terminal residue.
In this case the ordered regions would be in the "interaction with ESR1" domain (1968-2338), which as you can see has been solved experimentally (PDB 4D0N), and the sequence spanning from 965-1431 (part of it has been solved, PDB 2DRN). What I mean by "spaghetti" is basically: render these two regions then draw curvy arabesques/lines between, before and after them to show it is disordered (starting from their N-terminal or C-terminal).
Such a cartoon is best done in a program that is shows it utterly differently —to not confuse the viewer as this is for display purposes only. Namely, render the parts in PyMOL, but then add the connecting lines in Photoshop or Powerpoint or MS-Paint/Preview.