[gmx-users] LIE precision and accuracy

[chris.neale at utoronto.ca]

In spite of the fact that I have never used g_lie, I have used a  
variety of different techniques to calculate binding free energies,  
and I would be astounded if it was possible to get the precision that  
you desire with only 1 ns of sampling per state. Are you sure that the  
reported imprecision is different from the actual imprecision obtained  
via 1 ns of sampling? Generally, one evaluates the convergence by  
block averaging or some other convergence indicator. If you do find a  
measure by which your precision is reported to be very good, then I  
would suspect that your sampling is trapped and that your free  
energies are not accurate in any event. In short, I think that your  
comment about "rendering the energy difference meaningless" is  
actually quite astute.

Chris.

-- original message --

Hello,

       I am in the middle of using Gromacs 3.3.3 to evaulate  
interaction energies between a protein and two separate ligands to  
compare their binding affinities. In doing this I have run each  
solvated complex through a 1 ns MD simulation and each solvated ligand  
through a 1 ns MD simulation. After the trajectories were complete, I  
ran g_lie on the trajectories  for each case.  Next, I subtracted the  
average ligand LIE from the average complex LIE for each case to get  
the interaction energy . Data is as follows:

  Ligand 1 Ligand 2
LIE (kcal/mol)   -5.7   -8.9
St dev.    6.7    4.7


While the average energies themselves are reasonable, the st dev's  
seem to me to be too large - rendering the energy difference  
meaningless. I calculated the st. dev's based on the trajectories'  
g_lie values (in this case 1000 data points for a 1 ns run). I read in  
the tutorials that PME condition should not be used with g_lie, so I  
switched to cut-offs for my electrostatic forces. Is there anything  
else I might be missing? Again, I know the st. dev's are just too high  
to make g_lie useful so I blame myself for the initial setup. My .mdp  
file is below.

Thanks, Marc

; mdrun for C4AHL on unix parameters
title = trp_drg MD
cpp = /usr/bin/cpp ; location of cpp on linux
constraints = all-bonds
integrator = md
dt = 0.002 ; ps!
nsteps = 500000 ; total 1000 ps
nstcomm = 1
nstxout = 500 ; output coords every 1 ps
nstvout = 0
nstfout = 0
nstenergy = 500
nstlist = 10
ns_type = grid
pbc = xyz
rlist = 0.9
coulombtype = cut-off
rcoulomb = 1.2
vdwtype = cut-off
rvdw = 1.4
;fourierspacing = 0.12
;fourier_nx = 0
;fourier_ny = 0
;fourier_nz = 0
;pme_order = 6
;ewald_rtol = 1e-5
;optimize_fft = yes
; Berendsen temperature coupling is on in five groups
Tcoupl = berendsen
tau_t = 0.1 0.1
tc-grps = protein non-protein
ref_t = 300 300
; Use Energy group monitoring
energygrps = protein C6L OHY SOL Na+
; Pressure coupling is on
Pcoupl = berendsen
pcoupltype = isotropic
tau_p = 1.0
compressibility = 4.5e-5
ref_p = 1.0
; Generate velocities is on at 300 K
gen_vel = yes
gen_temp = 300.0
gen_seed = -1
free_energy = yes