Change number of iterations in mbar

[ncheron]

Hi,

I am using flamel to analyse FEP results (solvation free energy of acetate performed with Gromacs). With pymbar3.0.5 I encounter this issue choderalab/pymbar#320. After switching to pymbar3.0.3, flamel works for switching off the charges, but not when I switch off van der waals (I still have the same issue). I have read here choderalab/pymbar#419 a workaround (from Michael Shirts on March 22nd). Can you please guide me on where I should add these three lines in flamel to check if it works?

Thank you

Nicolas

PS : BTW, thank you for this tool, it is very useful

[orbeckst]

@ncheron at the moment nobody is working on flamel so I am afraid, you're on your own figuring things out.

If you want to contribute code and work on it then you're more than welcome!!

[ncheron]

Thank you for your answer. After switching from Gromacs2019.4 to Gromacs2021, my problems have disappeared and I don't know why. I will keep using Gromacs2021 for FEP.

[orbeckst]

FYI, alchemlyb 0.5.1 fixed that (but not in flamel yet I think)

[schlaicha]

I also just stumbled over that issue again - in flamel we just call the standard constructor of alchemlyb's MBAR introduced in alchemistry/alchemlyb#162 and changing the call in flamel/estimator/mbar.py to

mbar_est = alchemlyb.estimators.mbar_.MBAR(method="BFGS")

or

mbar_est = alchemlyb.estimators.mbar_.MBAR(method="adaptive")

both work fine in line with the discussion in choderalab/pymbar#419. So the passing of the arguments in alchemistry/alchemlyb#162 works as expected and the issue (at least for me) seems to be that in alchemistry/alchemlyb the default is method='hybr' which is not working for my particular system (roughly 1 out of 10 cases).

@orbeckst should I address this in alchemistry/alchemlyb? If sticking to method='hybr' I suspect that the problem is that in choderalab/pymbar the maxfev for scipy.optimize.root would need to be adjusted.

[orbeckst]

Better discussed in alchemlyb — you can raise an issue to change the default setting of the MBAR estimator.

[orbeckst]

@xiki-tempula is this issue addressed in the flamel rewrite, namely can one choose to use AutoMBAR for the ABFE workflow from the flamel command line?

[xiki-tempula]

@orbeckst So for alchemlyb 1.0, using MBAR defaults to use AutoMBAR.
for alchemlyb 2.0 (alchemistry/alchemlyb#268), The MBAR will defaults to use the robust method, which is the same as AutoMBAR.

Either way, we won't have this issue.

[orbeckst]

Thanks for the clarification, I am closing the issue.