User login

You are here

How can we obtain more information from protein structure?

We know - or believe - protein function is determined by structure. Crystallographic and NMR studies can provide protein structures with atomic-level details at equilibrium. MD simulations can follow protein conformational changes in time with fs temporal resolution in the absence or presence of a bias mechanism, e.g., applied force, used to induce such changes. However, often times only a very small amount of information is extracted from such massive amount of data; therefore only limited insights are obtained. Also, often times the process of extracting information from structure involves a lot of guess work, quite empirical, and hard to automate. How can we do better? Can mechanics help?

Comments

Ju Li's picture

A systematic way of coarse-graining the "massive amount of data" from MD or other discrete-agent simulations may define a new kind of mechanics. A 2003 PNAS paper by Miyashita, Onuchic and Wolynes indicates the importance of global views of proteins, more than just atoms vibrating locally. There are fundamental conceptual issues, such as whether stress is well-defined when there is long-range transmission of force, that need to be worked out. We need additional coarse-grained concepts or vocabulary (new mechanics) if existing concepts or vocabulary happens to be not applicable.

The notion of structure implies stable conformation.  Global conformational changes probably take place as rapid transitions among stable conformations.  Different stable conformations endow a protein with different properties and enable it to accomplish different functions.  Coarse-graining is one way to reduce representation.  Some of the coarse-graining methods, e.g., normal mode analysis, may be useful to identify pathway(s) of conformational transition(s), which must occur at frequencies much lower than those of the atomic-level vibrations.  At the extreme end of reduction of representation, stable conformations can be modeled by state variables, e.g., probabilities for the protein to assume one of the admissible states.  The dynamics can be described by a set of master equations governing the state transitions.  To elucidate the rate parameters and their dependence on external variables (e.g., applied force), however, may require local information with atom-size spatial resolution and atom-vibrating level temporal resolution.

Zhigang Suo's picture

I have a simpler question.  What are the available software to simulate dynamics of biomolecules?  How do I get them.  What is your own experience in using them?  If people with experience leave comments here, we ought to have a good picture of what is available, and perhaps we can have a better idea what is missing.

Xi Chen's picture

Zhigang, you can get CHARMM from across the street of your office. It's developed by Martin Karplus at Harvard Chemistry Department and one of the most widely used codes (I often heard people saying that Martin Karplus may get a Nobel Prize for his contribution in computational biophysics). It's open source and you can get it for $600, and it runs on Linux/Unix. There are some debates about the CHARMM forcefield since some people do not agree the simplifications they made (e.g. on the hydrogen bond) but overall the CHARMM developers argue the code works well for biomolecules. In our group, we also use Materials Studio: it's a commerical and greatly expanded version of CHARMM which has a much better interface, more comprehensive forcefield (so that we are not limited to biomolecules), and it can run under Windows -- of course you pay the price of losing much control, no parallel computing and kind of slow, and the license fee is much much higher too. Besides, the language (grammar) used in Materials Studio is awkward and its forcefield for metals is problematic. Both CHARMM and Materials Studio are MD simulation softwares which means that there is a long way to go to implement coarse-grain models into the codes. At this moment, we are working with a group of CHARMM developers to couple both FEM and coarse-grain models (as subroutines) with the CHARMM, to study the concurrent multi-scale mechanics of proteins.

Hi, Prof. Suo.

For simulation of biomolecular dynamics (e.g. protein dynamics), as you already knew, molecular dynamics (MD) simulation is widely used. In my experience at my Ph.D. work, at that time, I utilized the TINKER molecular modeling package for protein dynamics. TINKER is free-ware that is uploaded at http://dasher.wustl.edu/tinker/. TINKER typically used the CHARMM potential field, developed by M. Karplus at Harvard.

Except CHARMM potential fields, there are several empirical potential fields such as AMBER, OPTLUS, etc. Most importantly, the low-frequency normal modes of protein structures, which dominates the protein dynamics, were known to be insensitive to the details of potential fields. This suggests that protein structure (topology) plays a more important role than the potential field does.

You may also use the normal mode analysis (NMA) for dynamic behavior of biomolecules (e.g. protein) and/or molecules. NMA employs the empirical potential fields such as CHARMM, AMBER, etc. to calculate the stiffness matrix at equilibrium position. But I do not have any idea about commercial or free-ware software for NMA.

 

We use the NAMD program developed by the Schulten group, which is also free.  See Ref. 22 of the paper on my other posting.

But a more important issue is what kind of "new mechanics" do we need? 

It's unimaginable that anyone would calculate the dynamics of all the molecules in a volume of gas to determine its flow. We use fluid mechanics instead. Similar comments can be made for molecules in a solid and elasticity.  What would be the equivalent theory for proteins?

Zhigang Suo's picture

Cheng:  I'm fascinated by the questionn that you've raised.  It seems that you are looking for a model or, rather, an approach to set up a model for any protein of your choice.  The model can take input from any source (computation or experiment), and can predict something useful to someone. 

Now, a question for you, as well as for anyone with experience in this, could you please point out any existig work that invents models of this type?  I'm particular interested in models that help to relate experiments and predictions.  Metadynamics pointed out by Ting Zhu miight be an example, which I'll check up.

Ting Zhu's picture

I think "metadynamics" is a very promising coarse grained modeling approach, which was developed by Parrinello and his coworkers. This method focuses on the long time evolution of some carefully chosen collective variables. Anyone has more experience on this method?

Zhigang:  I am more problem-oriented than method-oriented.  So I will leave your question to others who are more knowledgable.  We use methods we come across in order to solve problems in hand and complain about the inadequacy of these methods.  But seriously, this is an area that, I believe, mechanicians can contribute.

Let's examine some fundamental concepts of mechanics.  Take, for example, two questions in Ju Li's post one at a time. Is stress a useful concept?  The textbook definition may no longer apply as it is hard to argue that portions of a protein satisfy the continuum assumption.  But proteins maintain their structures even when they are undergoing conformational changes.  So, motions of neighboring atoms are correlated.  From the trajectories of each individual atoms calculated via MD simulations, discrete strain and strain rate fields can be calculated, and for the most part, they are "continuous".  So using "finite element" to approximate the collective motion of a protein segment seem reasonable, which may be the basis for the approach taken by Xi Chen.  The correlated motions are results of inter-atomic interactions, as specified by the force field.  Inter-atomic interactions decay rapidly with distance, which may provide an opportunity to define some kind of "surface force" - the basis for the concept of stress.  The problem is that these interactions are specific to the atoms involved.  By comparison, the inter-molecular forces in a volume of gas are the same regardless of the particular molecular pair involved because they are the same molecules!

Subscribe to Comments for "How can we obtain more information from protein structure?"

Recent comments

More comments

Syndicate

Subscribe to Syndicate