search

OpenFreeEnergy / openfe

The Open Free Energy toolkit
https://docs.openfree.energy
MIT License
129 stars 17 forks source link

PyMBAR 4 convergence issues #443

Open IAlibay opened 1 year ago

IAlibay commented 1 year ago

We're seeing convergence issues on very innocuous transformations when using PyMBAR 4, e.g. lig_ejm_31 to lig_ejm_50 from the tyk2 set.

We'll need to dig into it and work out why this is happening.

For now we'll temporarily pin to <4 so we can get folks started.

mrshirts commented 1 year ago

Let me know if you can get a set of data to reproduce - I'd be happy to try to figure out what the issues might be. pymbar SHOULD be more robust in all cases, so I want to know what corner cases are having problems.

xiki-tempula commented 4 months ago

Is there any update on this? We have to stick with openfe 0.6.1 due to this pin.

mrshirts commented 4 months ago

I should have time relatively soon to look at this, but would need a data set to figure out what was going on.

xiki-tempula commented 4 months ago

I think the robust method should solve this problem? Pymbar 4 has been very stable in our pipeline with zero failure due to this issue. In this issue (https://github.com/OpenFreeEnergy/openfe/issues/833), there is even a case where Pymbar 3 will fail but Pymbar 4 robust solves the problem.

IAlibay commented 4 months ago

We have a potential fix for this by optionally enforcing the robust solver, but it needs testing on a larger set of production simulations which unfortunately we've not had time to do yet.

Right now the best I can offer timeline wise is that it is a priority and we will try to get it sorted as soon as possible, but it may be after alchemistry.