Manual analysis of predicted models

Back to index

Protein viewers

Rasmol (Roger Sayle) - good for simple analysis of protein structures.
download    tutorial (Gale Rhodes)
You can try protein explorer if you like, but protein explorer is web-dependent and doesn't work on UNIX systems.  I still recommend Rasmol.

Swiss PDB viewer (ExPASy)- good for manipulating structures and performing calculations on structures, but the graphics are so-so and the program easily crashes on UNIX systems.
download    tutorial (Gale Rhodes)

Pymol - (Warren DeLano) best graphics, and beautiful figures are easily produced.  Very object-oriented, can have multiple selections running at the same time.  The syntax takes time to learn, but it is worth the time if you are interested in detailed molecular analysis. 
download   user's manual

Biological ranking criteria

There is no easy technical manual for this.  The best tools for viewing proteins are a good biochemistry/biophysics/protein structure background plus experience.

I consider the three most important criteria to be these:

1) consistency with biological information
It is rare to find a structure that matches your biological information perfectly.  However, the interface should be in roughly the area suggested by biological information, and it is preferable if at least some of the essential residues (if there are known to be any) are at the interface.

2) close packing:
Packing (van der Waals attraction) is one of the major forces governing the structure of proteins.  All real protein interfaces are superbly well packed, and you should filter models for good packing.  Some imperfection is to be expected for Rosetta models, but structures with large voids or a small contact area should probably be discarded.  However, we often get Rosetta models that are well-packed that still are wrong, probably because we have over-relaxed the side chains in order to achieve a false positive docking configuration.

3) radius of gyration
Generally, but not always, configurations with as small of distance between the centers as possible are more likely to be correct than those with larger distances.  One physical argument for this is that the docking partners lose more rotational freedom in a complex with a high radius of gyration than in a complex with a low radius of gyration. 

4) Good specific interactions at the interface
These are difficult to assess but good specific interactions may be a sign of a good docked complex.

Watch out for what I call the "large interface problem."  Sometimes rosetta rewards large interfaces, even if the individual contacts are not very good or the interface violates biological information.  In addition, shape complementarity, while it is attractive, can be misleading.

I would say the number one rule is to collect as much biological information as you can ahead of time and to pay attention to it while analyzing structures.

Back to index