Hi everyone,
I’ve been using MBAR from the fe-toolkit package to reconstruct a 1D PMF, aiming to validate an issue I encountered with my input when analyzing a reaction free energy profile from a QM/MM umbrella sampling simulation reconstructed with WHAM.
The collective variable (CV) I used is a linear combination of two distances with different weights — an antisymmetrical distance (ASD) with weights scaled as 2 and -1, to reflect a two-step reaction mechanism.
According to the MBAR documentation, the force constant unit should be kcal/mol·(unit)². I therefore kept the same biasing constant as I used in WHAM (1/2, corresponding to the Amber disang file). However, I was surprised to see that MBAR produced free energy values roughly 10 times higher than expected — both in terms of expected values and compared to WHAM (grossfield) results.
When I rescaled the force constant by a factor of 1/5, the results became much more plausible. This makes me wonder: is MBAR particularly sensitive to the exact definition or scaling of the CV? More specifically, does MBAR inherently not support antisymmetrical distances, or is there a recommended way to treat such CVs?
I’d greatly appreciate any insight or references where I could learn more about these behaviors.
Thank you in advance!
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu May 01 2025 - 09:30:02 PDT
