Well, you might have a look at the MD literature over the last few years -- the system sizes have been steadily increasing, while at the same time making predictions and being compared with quantitative experimental data. At the same time the list of known issues with MD has been steadily shrinking (people in the field do know and acknowledge that there are still some areas where the simulations fail, but a lot of the problems have already been worked out). The simulation being reported here isn't even the biggest MD run that has ever been done; that title would belong to Sanbonmatsu's ribosome work or Namba & Kitao's flagellar filament simulations. So this certainly doesn't seem like it is coming out of nowhere, as you might be concerned -- there have been a lot of other big simulations done recently in the 100,000 atom and up range.
The simulations for this study were done using smoothed particle mesh ewald, which lets you capture all long range interactions (including those in the local periodic cell) in the fourier space part of the calculation. You can have a look at Phillips et al., JCC 26(16):1781-1802 (2005) and its references for more on the method, including the necessary approximations and its shortcomings.
Honestly, I think you're being a bit harsh on the general topic of MD force fields. Now lets be clear: all of the forcefield stuff is done working under the assumption that there are no changes in electronic structure, and that instead interactions between "bonded" atoms stay close enough to their equilibrium structure to be treated with a harmonic approximation. Working within those guidelines, though, you DO derive a forcefield from ab initio QM work, and then that's it -- you don't tune it for your particular case. You can choose from a short list of parameter sets, but these are relatively few in number, and in general one group applies the same parameters to all the problems they deal with; it isn't like people are trying out CHARMM, Amber, OPLS, and everything else under the sun to see what gives them the best results. You say "of course it is going to be realistic", but isn't that the point? Lets take a more coarse example: people these days use computer simulations of windflow around a potential airplane design to test how it is going to perform. I can assure you that they are not doing ab initio simulations on all of the constituent molecules of the air; they're probably not even tracking the individual molecules, because it isn't appropriate to the problem. Yet I doubt you'd say that they aren't modeling something. MD is the same way; it is a set of approximations appropriate for a certain subset of problems (and you can do a literature search in your favorite place to see how many applications it has already had), and represents our best current techniques for balancing the tradeoffs between detail and computing requirements for watching the mechanical motion of proteins that are not involved in chemical reactions. It certainly isn't perfect, but it is constantly improving.
Slashdot Mirror
Well, you might have a look at the MD literature over the last few years -- the system sizes have been steadily increasing, while at the same time making predictions and being compared with quantitative experimental data. At the same time the list of known issues with MD has been steadily shrinking (people in the field do know and acknowledge that there are still some areas where the simulations fail, but a lot of the problems have already been worked out). The simulation being reported here isn't even the biggest MD run that has ever been done; that title would belong to Sanbonmatsu's ribosome work or Namba & Kitao's flagellar filament simulations. So this certainly doesn't seem like it is coming out of nowhere, as you might be concerned -- there have been a lot of other big simulations done recently in the 100,000 atom and up range.
That's fair enough; your comments elsewhere had lead me to believe you were more critical of the technique as a whole.
The simulations for this study were done using smoothed particle mesh ewald, which lets you capture all long range interactions (including those in the local periodic cell) in the fourier space part of the calculation. You can have a look at Phillips et al., JCC 26(16):1781-1802 (2005) and its references for more on the method, including the necessary approximations and its shortcomings.
Honestly, I think you're being a bit harsh on the general topic of MD force fields. Now lets be clear: all of the forcefield stuff is done working under the assumption that there are no changes in electronic structure, and that instead interactions between "bonded" atoms stay close enough to their equilibrium structure to be treated with a harmonic approximation. Working within those guidelines, though, you DO derive a forcefield from ab initio QM work, and then that's it -- you don't tune it for your particular case. You can choose from a short list of parameter sets, but these are relatively few in number, and in general one group applies the same parameters to all the problems they deal with; it isn't like people are trying out CHARMM, Amber, OPLS, and everything else under the sun to see what gives them the best results. You say "of course it is going to be realistic", but isn't that the point? Lets take a more coarse example: people these days use computer simulations of windflow around a potential airplane design to test how it is going to perform. I can assure you that they are not doing ab initio simulations on all of the constituent molecules of the air; they're probably not even tracking the individual molecules, because it isn't appropriate to the problem. Yet I doubt you'd say that they aren't modeling something. MD is the same way; it is a set of approximations appropriate for a certain subset of problems (and you can do a literature search in your favorite place to see how many applications it has already had), and represents our best current techniques for balancing the tradeoffs between detail and computing requirements for watching the mechanical motion of proteins that are not involved in chemical reactions. It certainly isn't perfect, but it is constantly improving.