0
Posted October 2, 2013 by Dr. Henri Montandon in brain networks
 
 

AD-ENM server: computing with elastic network models

cool runnings 61 titlecool runnings 61 preferred pix

 

 

 

 

http://enm.lobos.nih.gov/

Modelers run into trouble when very high degree of freedom models need to be calculated. Not many of us have supercomputers to use, or even access to one. So theoreticians are always on the lookout for ways of simplifying their models, preferably to the size where they are PC compliant.

Computational intractability can put even relatively simple neural network models on the rocks. For example, how would you characterize, computationally, a 12-node network where each node is connected to only one other?

It turns out that in biology, these kind of network characterization problems have been in play for many years. Biologists stumbled over them when considering one of the most interesting, and intractable, problems in modern structural biology: the protein folding problem.

A protein, which is basically a folded polypeptide chain, can be simply modeled as a mass-spring system. Elastic network  models have been developed to explore the relation between protein structure and biological function. (If this sounds familiar, it should. Much of the discussion of neural reentry, for example, rests on the quest for the relations between patterns of cortical signals and neurobiological functioning.)

The illustrations below depict a coarse-grained elastic network model compared with a ball-and-stick molecular model.

Computations for the ball-and-stick model require a supercomputer, while information-rich simulations of the elastic network can be run on a PC. In the elastic network model, the nodes are connected by linear elastic springs. Here, the nodes are a carbon atom at the center of a given amino acid. Because of the recursive nature of protein structures, symmetries can further constrain the model, significantly reducing the degrees of freedom and improving computational efficiency. These simplifications, and others it is not necessary to review here, allow easy modeling access to the global dynamics.

cool runnings 61  molecular vs elastic

 

 

 

 

 

 

 

 

 

 

 

The good folks at the Computational Biophysics Section of the Laboratory of Computational Biology have made available a free elastic network model server that will calculate for you some of the things that make these models so valuable.

For example, you want to study the conformational changes of a protein molecule you are interested in? AD-ENM will do it for you. Feed in a few parameters, and faster than you can say dynamic molecular model, those changes will appear on your screen.

Suppose, however, your interest is not molecules, but neural networks? Molecular parameterization in elastic network modeling has facile analogies with parameters in neural network models.

Nobody in the neuroscience world has used elastic network models to study neural networks. Yet here is a computationally efficient and simple modeling scheme that can be used to compare modes and changes of modes among networks. Just the ticket for neural network study. Tell your friends.

 

 

 

 

 

 

 

 

 

 

 


Dr. Henri Montandon