Differences between MBAR and BAR

[AnguseZhang]

When performing FEP, I found there are significant differences between MBAR and BAR, as the figure shows.

I used example from FESetup: T4 with benzol and oxylene. The large gap often appears in solvated simulations. (I would also want to know the reference value of complex and solvated ddG
for this system)

When only putting data of two adjacent states into MBAR, MBAR results are the same with BAR.
But putting data of all states, the gap appears.

This is my code

python alchemical_analysis.py -d . -q xvg -p dhdl -u kJ

This is original file.

data.zip

I also tried alchemlyb and using original pymbar, but the results are the same.

In my view, MBAR and BAR are both unbiased estimators, and thus the estimation (free energy difference ) should be the same. Could you please give me some instructions ? Thanks !

[davidlmobley]

BAR and MBAR ARE slightly different, so will yield slightly different results. Sounds like your question is best directed at the pymbar issue tracker though.

Marking as closed since it's not an issue with this package.