0.25.0 - Moved multistate and mpi modules and new trailblaze algorithm

API-breaking changes

• The yank.mpi module and the objects in the yank.multistate package, which were deprecated in 0.24.0, have now been removed, and they can be found in the mpiplus and openmmtools libraries respectively.
• The function yank.pipeline.trailblaze_alchemical_protocol has been renamed yank.pipeline.run_thermodynamic_trailblazing (#1180).

New features

• The thermodynamic trailblazing algorithm used for the authomatic generation of the alchemical path is now capable of resuming after an unexpected interruption or crash. The samples generated during the process are used to initialize the replicas of the replica exchange or SAMS free energy calculation. This behavior can be controlled through the start_from_trailblaze_samples YAML option (#1176, #1180).
• It is possible to control more options of the thermodynamic trailblazing algorithm and to discretize an alchemical path given through mathematical expressions enslaved to a generic variable (#1180).
• Added a --setup-only flag in the yank script CLI command to run the automatic setup pipeline without running the free energy calculation (#1178).

Bugfixes

• Fix a bug in which a list of experiments: [exp1, exp2] in the YAML file containing an unknown experiment name would fail silently without error (#1178).
• Fixed a problem that would prevent YANK to work with Cerberus >= 1.2 (#1180).
• Fixed the error message displayed when the anisotropic dispersion correction cutoff was too big for the box dimension (#1181).

Enhancements

• By default, the automatic determination of the alchemical path now starts with the harmonic/flat-bottom restraint turned off and activate it in intermediate states instead of keeping the restraint activated throughout the calculation and reweighting in the analysis stage (#1176).

Known issues

• Using parallel MPI processes causes poor mixing of the odd thermodynamic states while the mixing of the even states is normal. We're still investigating whether the issue is caused by a change in the MPI library or an internal bug. For now, we recommend running calculations using only 1 GPU (choderalab/openmmtools#449 and #1130).
• Simulations restored from a checkpoint file have their velocities reset to zero (#1115).
• Forward and backward convergence analysis free energy traces in the Jupter notebook are incorrect (#971).
• Setup will fail if .mol2 atom substructure ID matches filename (#703).