[gmx-users] PMF Transmembrane proteins

[Nash, Anthony]

Dear gmx users,

I posted a couple of weeks ago with regards to correctly using umbrella sampling and the WHAM on atomistic transmembrane proteins with a reaction coordinate as a function of interhelical distance. I have a single TM dimer, but with a different transmembrane domain face at the helix-helix interface. 

What did I conclude from the replies? Correct calculation of the differences in free energy is taken by normalizing to zero at the plateau (flat) of the graph i.e., where your PMF graph (after g_wham) flattens out you apply the -zprof0 at this point, before calculating the difference between of each system. 
 
The problem I face is that this occurs around 6.8 - 7.5nm along the reaction coordinate. And, applying a umbrella window every half angstrom has made this extremely computationally expensive. However, I have persisted, and I have all my graphs, and I have a plateau on all of them. Yet the PMF graphs between 4 - 7.5 nm are not yet converging or close to the same sampled energy, even though all four systems are identical in amino acid composition. Each window has ran for up to 40 ns. 

So when I normalise, depending on where I normalise I will get massively different difference in free energies. 

I have tried the strategy of taking the windows of a particular systems at 4.5 nm to 7.5 nm along the reaction coordinate (where there is no short range interaction), and applying those windows into the other three systems. This brings their region of plateau a lot closer together, whilst preserving the windows and the convergence of the region along the reaction coordinate where there are close-range forces in play (converges 0.8 nm - 4.5 nm).

Alternatively, could I make the assumption that as their amino acid composition is identical across all four systems, I can normalise where each of the four graphs finish converging (around 4.5 nm)? 

I would really appreciate any advice from someone who has insight on this issue. I have looked through many journals and founds lots of entries where they just stop at 2.0 nm with no plateau, yet I haven't found a single WHAM of reconstructing umbrella windows where they normalise at the plateau. I have also contacted a few authors of journals with regards to why they cut off at 2 nm. None of them took into consideration that their PMF graphs were still raising. 

Many thanks
Anthony