[gmx-users] Higher that usual salt bridge occupancies in Amber99sb*-ildn forcefield despite adding nonbonded corrections

[Priyesh Mohanty]

Dear all,I've been using certain non-bonded corrections to the strength of salt bridges in this forcefield based on a recent publication for the last 6 months or so. The following correction was added to the ffnonbonded.itp file which increases sigma from 0.310 to 0.318 : 

[ nonbond_params ]N3                       O2    1  0.318496        0.79054352138259915032 ; +0.08 for LYS
N2                       O2    1  0.318496        0.79054352138259915032 ; +0.08 for ARG
While these corrections initially led to a drastic improvement in the kinetics of various systems simulated, I've now noticed an increase in the strength (~25% increase in occupancy) of salt bridges over the last two months or so despite having the corrections applied. It is not as high as it is without the correction, but noticeably higher. It is having a significant effect on the kinetics of some of my systems and I cannot reproduce the results of my previous runs with the same corrections applied. I've carried out both cpu only and cpu+gpu production runs on gromacs 4.6.7 and 5.1.2 on multiple machines (both workstations and a cluster) using the verlet cutoff scheme but the increase in occupancy seems to be consistent across all of these runs. The problem arose at the same time simultaneously in all of these machines back in January earlier this year. I am quite puzzled by this and would really appreciate any suggestions regarding this.
Regards,Priyesh MohantyResearch ScholarNational Centre for Biological SciencesBangalore, India