- This topic has 3 replies, 3 voices, and was last updated 10 years, 1 month ago by Anonymous.
October 14, 2013 at 3:33 pm #1727Anonymous
I’m a newbie at Rosetta, and I need to figure out if i can do something.
I have an assymetric homo dimer PDB structure, but each monomer has about ~230 missing residues (with predicted high probability of being disordered). Each monomer has this organization (from amino to carboxyl):
[known structure domain][missing structure domain]
The protein-protein interactions of the asymm dimer are between the domains with known structure. I need to figure out how the missing structure domains interact with the structured ones, through the generation of thousands of models.
At a monomer level, I was thinking about modelling the complete sequence by ab initio. Then, using as templates the PDB of the known structured domain plus the structure of the ab initio modelled missing region execute a comparative modelling to obtain the complete monomer structure.
The question is, is there a better way to make a mix of ab initio plus comparative modelling?
is it posibble to implement this modelling at assymetric dimer level? specially considering that the missing structure domain of one monomer perhaps may interacts with the other monomer.
Any help or suggestions are greatly appreciated!!
October 14, 2013 at 3:48 pm #9415Anonymous
From my experience, Rosetta’s energy functions are all built for mainly structured regions. Some of us have discussed creating a scorefunction particularly for disordered regions, but nothing has happened yet.
Is there any data that suggest the 230 residues become ordered via some transition/phosphorylation/etc?
Are there regions within the 230 that are ordered, or the whole thing is predicted to be highly disordered?
Honestly, If it’s the whole 230 residues, I would not have much faith in anything created via homology modelling or ab-initio. Someone on here may think differently, but both will give order to that 230 residues.
Are there any Secondary Structural elements predicted in the domain? If so, then ab-initio may give you something, but again, its how much faith you put into the resulting models. If you want to get an idea of what is possible, then certainly ab-initio may be cool. If your domain was smaller, I would suggest floppytail…
If there is something that has a high degree of homology, then you could build the model via homology, and then run loop modelling on the loops…
October 14, 2013 at 8:08 pm #9419Anonymous
Thanks for your comment.
This 230aac have multiple Tyr and Ser/Thr phosphorylation sites. But there is only one PDB tha includes 5aac of this region with his phospho-Tyr interacting with another protein, nothing else.
The whole region is predicted to be disordered (IUPRED,PSIpred). In previous attempts of ab initio modelling this 230aac alone, some regions adopted alpha-helix conformation, with no more than 3 turns.
Unfortunately, all these 230aac are only conserved between some members of the family of the protein i’m studying.
But, ….beyond this fact of disorder, is there a way to model asymmetric dimers? May I can change the chain name “B” to “A” an pretend having only one chain?
October 15, 2013 at 2:25 pm #9421Anonymous
Asymmetric homodimers are modeled like heterodimers. Keep them as separate chains in the PDB, and don’t include any symmetry information. There may be some per-protocol specific things that you’d have to do to enable sampling of the rigid body degrees of freedom.
The problems you’d run into are those protocols which assume a single chain. (So can’t handle heterodimers in any fashion.) Abinitio and certain versions of homology modeling fall into this category. Sometimes you can fake it by pretending that everything is one chain, but often you’ll run into problems where they’ll want to connect up the N and C termini, probably where you don’t want to. The details will be protocol specific.
As Jared says, Rosetta does a poor job of modeling intrinsically unstructured regions, (if the secondary structure predictions are all loop, Rosetta’s fragment-based approaches won’t have much to work with) and will also do a poor job of modeling large structured regions of where there isn’t any additional structural information. For predicting structure in the 230 aa range, you really need to have a solved structure of a homolog, or at the very least experimental evidence (like NMR constraints, or some other distance constraints).
- You must be logged in to reply to this topic.