[gmx-users] Strange dV/dl profile in free energy calculation

[Robert Johnson]

Hello everyone,

Sorry if you get this twice. It seems to me that the previous message
didn't go through.

I'm computing the binding free energy of DNA bases (just the base, no
sugar, no phosphate) on a carbon
nanotube following the prescirption of the Dill group Wiki
(http://www.dillgroup.ucsf.edu/group/wiki/index.php/Free_Energy:_Tutorial)
and the Boresch paper (J. Phys. Chem. B., 107, 9535, 2003). I split my
calculation up into two parts. In the first part, I turn off all the
charges in my DNA base (guanine in this case). In the second part, I
turn off all the LJ parameters. I run MD at each lambda value for 15
ns. An image of the dV/dl plot obtained from the previous
simulations is located here:
http://dept.physics.upenn.edu/~robertjo/public_files/gmx/dvdl.jpg The
error bars are included in the graph (they are very
small...less than 1 kJ/mol in all cases).

My concern is the dV/dl curve for turning off the LJ parameters (i.e.
making the base "disappear") looks fairly strange. First it increases
and then decreases to a minimum value at lambda=0.8, and then
increases again. This looks significantly different than the identical
curve for disappearing a methane molecule in water:
http://dept.physics.upenn.edu/~robertjo/public_files/gmx/dgdl_methane.jpg
The methane calculation followed the steps discussed in the Dill group
Wiki and agrees with the result reported there. Is there anyone who
has some experience performing these calculations that can comment on
my curve for the DNA base? Does my curve look reasonable?

Thanks,
Bob Johnson