3D Protein Model - Suggestions for Programs

Question

[andrebolerbarros]

Hello everyone,

I have a GTPase protein that I'm interested in that I would like to predict its 3D model. Does anyone have a suggestion of program to use?

Thanks in advance

Answer 1

[Mensur Dlakic]

Modeling protein structures is a non-trivial problem, especially in the absence of highly homologous structures. It is particularly challenging for large and multi-domain proteins. There are many automated servers, so you will always get some some kind of structure if you submit your protein sequence to one or many of them. It is important to understand the value of a model and objectively evaluate how good it is. Ultimately it is up to you how to use the model, but some of it may be more or less justified. For example, a model based on >70% template identity and with good stereochemistry can potentially be used for many other downstream experiments such as molecular replacement, docking, mutagenesis, surface charge distribution, etc. A model that based on a template with <30% identity may be good for visualization, but care should be exercised in other applications.

I am going to assume that you have a template for modeling since GTPases are well-studied by structural methods. I suggest you first consult these resources:

• https://en.wikipedia.org/wiki/Protein_structure_prediction
• https://en.wikipedia.org/wiki/Homology_modeling
• https://en.wikipedia.org/wiki/List_of_protein_structure_prediction_software

Specific tools I recommend for homology modeling:

• https://www.nature.com/articles/nprot.2008.197
• https://swissmodel.expasy.org/
• https://modbase.compbio.ucsf.edu/modweb/

For targets with low or no homologous structures:

• http://robetta.bakerlab.org/
• https://zhanglab.ccmb.med.umich.edu/I-TASSER/

You will see that many of these programs can be downloaded for local use, but it is not easy to learn how to use them without guidance. There are many tutorials and review articles about protein modeling if you want to invest time.

[jgreener]

I would just add that the available methods that performed best in the absence of homologous structures at the recent CASP14 experiment are trRosetta and Zhang servers (I-TASSER and Quark). All are available as servers, though I don't know if they are fully updated to the most recent versions yet.