Cihan Aydin wrote: > I have two structures for a multidomain protein (in different > conformations) from the different genus in the same family. The system I > am looking is also in the same family but different from the two I > mentioned above. I have the individual structures for each domain for my > system but I don't have the full length structure for my system. I would > like to model the individual domains onto the two full length structures > from the other genii.
You should be able to do this with an alignment like
domain1 struc AAAAAAAAAAAAA------------------- domain2 struc --------------BBBBBBBBBBBBBBBBBB model aaaaaaaaaaaaa/bbbbbbbbbbbbbbbbbb
But since Modeller won't have any information about the interface between the two domains, they will be unconstrained and almost certainly won't come out in the correct conformation. One possible solution is to measure distances between nearby residues in one or both of your multidomain structures (if the interface region is similar) and add them as distance restraints to the corresponding residues in your model.
> In addition, I also want to model the > substrate-bound forms of these domains from my system to the full length > structures from the other genii, but the full length structures in > either case does not contain the substrates. I have x-ray structures for > every one of them.
If there is any kind of conformational change upon binding, Modeller won't be able to account for that.
> Also if I read it correctly, Modeller optimizes the structure with MD > using the CHARMM potential function in vacuo. Is there a way to > incorporate the AMBER potential into Modeller and do optimization in > solvated form (e.g. in TIP3P shell)?
In principle you could replace the CHARMM potential files (top_heav.lib and par.lib) in Modeller with the AMBER potential - it just has to be in CHARMM format. But I don't know what this would buy you, because this is essentially just used for amino acid stereochemistry, which is pretty well satisfied in most modeling runs.
Modeller does not incorporate solvation because this is implicit in the template structures, and would be very computationally expensive. You'd be better off using a dedicated molecular mechanics package for that sort of thing.
> Finally, I was reading the Eramian et al. paper titled "How well the > accuracy of comparative protein structure models be predicted?". Is the > trained SVM available somewhere to predict the accuracy?
The system is currently being updated to take advantage of more recent crystal structures and models, so is not currently available.
Ben Webb, Modeller Caretaker