Protein dynamics.



Both David and Konrad had several useful and thoughtful comments some of which are given below. I am still thinking about them.
 
To a first approximation David is saying we need faster computers, while Konrad is saying we need more experiments.
 
Even a small 100 residue protein (approx 1700 atoms) needs approximately 6000 waters (18000 atoms) to appropriately solvate (periodic box, with minimal switched cutoffs so as to avoid protein-protein interactions) the protein in an all atom Amber MD simulation for 100+ ps ...
 
The TIP3P water model was equilibrated at 300 K and 1 atm. In the literature I see it used at temperatures below and above 300 K. In fact it is used in non-isothermal simulations.
 Is the TIP3P water model ok to use for (isothermal) MD at 400 K and 1 atm ?
 
Also various numbers are used as the minimum distance between solvent and protein atoms (when adding water). I use 2.3 A.
 
With respect to Konrad's comments. I had thought the frictional co-efficient was a constant in LD and it was chosen based on the water-protein interface. In interesting that it is (possibly) the intra-protein contacts that dominate. Although I would think this would scale with the size of the protein and therefore the frictional co-efficient would be protein dependent (or at least have two terms taking into account the water-protein and protein-protein contacts and these in turn would change during a simulation as the protein conformation changed increasing or decreasing the solvent accessible protein surface).
 
I think the key point here ("For slow large-scale motions, you can leave out electrostatics completely, assuming no unfolding of course ...") is what is considered a structural change vs unfolding.
 Thanks again for your thoughtful comments.
 Laurence
 
 On Mon, Jun 06, 2005, Laurence Lavelle wrote:
 > By protein motion I mean a folded protein in a constant temperature
 > simulation (for example with a distance dependent dielectric, no
 > electrostatic cutoffs, no periodic boundaries) and comparing the protein
 > dynamics at different temperatures.
 With the above conditions (no solvent, distance dependent dielectric) there is
 little reason to expect that either the average structure or the dynamics you
 see in the simulation will be realistic.  If you want to get any reasonable
 account of protein dynamics, you need to be looking at MD simulations in
 explicit solvent.  If you want good results as a function of temperature, you
 should take care that the water model you are using is known to produce good
 dynamical results vs. temperature for pure water.
 ...dac
 
 
 On 06.06.2005, at 23:43, Laurence Lavelle wrote:
 
 In looking at the dynamical motion of a protein, is Amber (Cornell et
 al. (1994) force field) (Amber 99 parameters) considered a reasonably
 realistic protein force field (or, as good as or better than most) ?
 In looking at the dynamical motion of a protein (for example using
 Amber), what are the pros and cons to doing Molecular Dynamics vs
 Langevin Dynamics vs Monte Carlo ?
 
 It all depends on what dynamical quantities you wish to calculate, and
 on which time scales.
 If you look at long time scales for a protein that has a single stable
 conformation, then you can get good results with very much simpler
 models at much lower cost. See
   K. Hinsen, A.J. Petrescu, S. Dellerue, M.C. Bellissent-Funel, G.R.
 Kneller
   Harmonicity in slow protein dynamics
   Chem. Phys. 261, 25-37 (2000)
 for an example. At the other end of the time scale spectrum, if your
 study involves quantum effects, no Molecular Mechanics model will be
 good enough.
 As for sampling techniques, note that Monte Carlo is not a dynamical
 technique at all (it has no time scale), even though in some situations
 a time scale can be assigned a posteriori. Langevin and Brownian
 dynamics add an implicit source of friction and thermal energy, which
 should correspond to some physical feature of the model, i.e. the
 explicit modelling should be less detailed.
 In this context, it is worth pointing that the major source of friction
 in a protein is not the solvent, as is frequently believed, but fast
 interactions in the protein itself. See the article quoted above for a
 demonstration.
 
 {Some will say it depends on the details (distance dependent
 dielectric vs explicit solvent water molecules, no cutoffs vs with
 switched or shifted cutoffs, with or without periodic boundary
 conditions, etc.). However I am
 
 Again this depends on the time scales. For slow large-scale motions,
 you can leave out electrostatics completely, assuming no unfolding of
 course, because on such a coarse length scale the charge density is
 practically zero. On the other hand, for localized events, a correct
 representation of electrostatics is very important.
 
  hoping (in addition to the above two questions) to get a general
 sense of how realistic MD, LD and MC are with respect to illustrating
 protein motion.
 
 We don't know that much about how realistic simulations of protein
 dynamics are. Experimental information is mostly on very small time
 scales or on time scales so long that they are out of reach of
 simulations. In between those two extremes, there is nothing that
 simulations could be tested against.
 Konrad Hinsen