Protein flexibility and elastic networks

Elastic networks and protein motion

Anisotrpic Network Model: Systematic Evaluation and a New Web Interface

Ivet Bahar is one of the pioneers of using elastic network models (ENM) for studying protein motions. A way of doing normal mode analysis (NMA), ENMs have become increasingly popular in the past few years due to the work of Prof. Bahar and others like Qiang Cui. The biggest value of ENM models and similar methods is to look at large scale cooperative motions of protein structures that are not accessible via my method of choice (molecular dynamics). Reduced representations like ENM have played a huge role (and will continue to do so) in making such calculations tractable. In this paper, Bahar and co-workers look at the performance of anisotropic network models (ANM) on a large, database, scale which they have implemented as a web server. This new server also focuses on a usable graphical interface for analysis and interpretation, something missing thus far among the many ENM servers.

The analysis presented has been performed on a set of 176 proteins. Solvent accessibilities are calculated on a per-residue basis. The ANM is compared to experimental B-factors reported in the PDB files for the chosen proteins. The authors note that this is done, since B-factors are readily reported, but the ideal property to measure using ANM are coorperative motions, which normally require comparisons from more than one PDB structure. I found it interesting to note that traditional Gaussian Network Models (GNM) showed better correlation to experiment than ANM. I wonder if any changes can be made to the ANM to rectify that?

The fact that globular proteins showed the best correlation is not surprising. However, I was a little surprised at the variation. Again, I wonder if the results can be improved from a modification of the ANM? In general, the paper does a good job of examing the relative accuracy of various factors, including residue type and secondary structure. While some of the senstivity to certain factors is probably a cause for concern, overall the results are fairly encouraging. I certainly feel like there is an opportunity to perhaps add some knowledge-based features for improved results, at least as a filter of some sort.

One of the nicer features of the implementation is the visual interface. While the quality of the interface could be better (Jmol is not my favorite), overall the type of data available, interactivity and accessibility seems to be good enough (fig 9) and will likely make the server more popular than some others.

One concern I do have is that the database only includes monomers and proteins for which the monomer is the functional form. That is far too limiting in the long run and hopefully efforts to go beyond this limitation are under way.

Multi-resolution representation of protein flexibility

Hierarchical and Multi-Resolution Representation of Protein Flexibility

Continuing with this theme of protein flexibility comes a paper from Michel Sanner’s group at Scripps. I was especially intrigued by this since I have always associated Prof. Sanner with molecular graphics and docking. However in keeping with his groups skill at developing computational platforms, this paper describes a novel data structure, the Flexibility Tree, which is presented as a general platform for representing descriptors of molecular flexibility and multiple scales of molecular motion. I am a sucker for any such platform, so I had to give this paper a look see.

I am no expert on data structures, but conceptually the Flexibility Tree (or FT) is not that complicated. Each node represents rigid body motions of a collection of atoms (at varying levels of resolution). The root node represents either a macromolecular or an assembly. The relative positions of the nodes represents various collections moving relative to each other. There is a lot more detail included in the model, which makes it quite rich. I wonder if at some point, one can even add MD information to the model, to make the microscopic information content richer and perhaps even to correlate microscopic motions with macroscopic motions, a true exercise in multiscale simulation.

The key then is the partitioning of a system. Again, I am not sure if this is automated, but the whole structure is part of PMV and Vision (which seems to have superceded VIPer), a molecular graphic application and a visual programming environment respectively out of the Sanner group.

The FlexTree package is essentially in concept stage right now, but the authors do an excellent job of showing how it could work and the kind of information it could contain. The software skills of the Sanner group are top notch and I think with the right collaborations there is a lot of use that can come out of the FT approach and applications. It is important that the level of complexity not rise to a level that the average protein structure modeler feels completely at sea.