[gmx-users] Normal mode analysis in solvent: Hessian of protein only

[Rachel Baarda]

Dear All,

I am trying to find the full normal mode frequency spectrum of a small
protein. My system is composed of ~1900 atoms total, of which 130 are
protein atoms and the rest are solvent atoms. Since the Hessian matrix is
3Nx3N, the full spectrum for my system consists of ~5700 eigenfrequencies
(5700=3x1900). I am only interested in the frequencies stemming from
displacements of the protein atoms, not the solvent atoms. This would yield
a much smaller spectrum of only 390 frequencies. In other words, rather
than compute the full Hessian Hij=d^2U/(dxi)(dxj) for all (xi,xj), I want
to include only (xi,xj) coordinates belonging to protein atoms. Is there a
way to accomplish this?

I tried using gmx trjconv to remove solvent from my .gro and .trr files
prior to constructing the Hessian matrix, but this is not a good solution
because (1) my system is no longer properly energy minimized, which yields
negative eigenvalues and (2) I want to include the effects of solvent
rather than perform the analysis in-vacuo.

I also considered using implicit solvent, but I am using the Martini force
field. I understand that the implicit solvent adaptation for Martini is
currently available only for lipids, not proteins.

I appreciate any input on this - thanks in advance!

Rachel