A->B and B->A Transforms are Both Positive

We are trying to troubleshoot some issues using perses on some ligand transformations. We’re testing a single forward and backward transformation ( A-> B and B -> A) which is the most trivial closed cycle. Our intuition tells us that the magnitudes of these two transformations should be equal and opposite to each other, and this is what we see with the experimental data. Right now, the calculated ddGs are not adding up to zero, but instead are both very positive. Can you help shed some light on this? Could this be a units problem? This is what the mm_graph looks like:

OutEdgeDataView([(9, 10, {'calc_DDG': 5.18712727908637, 'calc_dDDG': 0.4801874381154462, 'exp_DDG': -0.9193053246474641, 'exp_dDDG': 0.0}), (10, 9, {'calc_DDG': 11.124917041483098, 'calc_dDDG': 0.4120190004511572, 'exp_DDG': 0.9193053246474641, 'exp_dDDG': 0.0})])

[jchodera]

@mikemhenry @ijpulidos : We should add some A->B and B->A consistency tests (likely just ligand in solvent, not complex) to our GPU CI as well. We can start with examples we already have, but can add a public example from @casebeerVRTX once provided.

[casebeerVRTX]

I wanted to follow up with this issue since it is now resolved. We ended up rebasing the lipid-version of perses to the most recent version, and changed the atom mapping strategy because the same ring was getting mapped incorrectly. This doesn’t solve the issue of both of these being positive, but we were able to find a work around.

Thank you!