Replica exchange with ML/MM

[pamazzeo]

Is your feature request related to a problem? Please describe.
I'm trying to use Sire for Hamiltonian Replica Exchange to decouple a solute from the solvent using a QM(ML)/MM electrostatic embedding scheme. I can successfully pass a callback function to Sire (via sr.qm.create_engine) to perform standard molecular dynamics with ML/MM potentials. However, when performing the solute-solvent decoupling (e.g. using decharge), the ML/MM engine must be aware of the current alchemical state, as the $\lambda$ parameter needs to scale:

• The ML/MM electrostatic interaction energy;
• The polarization of the ML region induced by the MM solvent.

Describe the solution you'd like
Is there a way to pass this parameter to the callback function that does the ML/MM prediction during the simulation?

Describe alternatives you've considered
Passing lambda as a fixed parameter to the callback function when setting up the different replicas is insufficient, as it does not account for the changes in alchemical state that occur during replica exchange.

[lohedges]

Hello there. Many thanks for your question. Are you able to share a code snippet so that I can get a better idea of what you are trying to achieve? From your description, it sounds like the fes-ml package might be suited, although this currently doesn't support replica exchange. What you want to do is probably easiest using the OpenMM-ML interface, where you can use emle_force, interpolation_force = engine.get_forces() with the latter force used to interpolate the embedding itself. Currently this approach doesn't support link-atoms, but it doesn't sound like you would need them in this instance.