Near term goals and planning

[pgrinaway]

This is what I am currently thinking--please let me know if you disagree.

perses has these parts which need development:

*TopologyProposalEngine - this already has an implemented base class and subclasses for small molecules. The small molecule bits need to be refactored for two reasons:

• The alignment can have multiple results, so the choice of alignment should be random and factored in to the acceptance probability. This is more urgent and I will do it soon.
• The innards should be refactored to use the OpenMM APIs rather than tleap. This is lower priority, since I think the priority is getting something that actually works. This also doesn't change the outward-facing API.
  • GeometryEngine - This is implemented, but needs a couple things, some of which are high priority:
  • GeometryEngine needs a suite of tests. This is very high priority and was the next thing I wanted to do.
• GeometryEngine should use OpenMM to calculate potentials. The code for this is already started, but is in the unmerged closed PR. Now that we have a set goal, I can finish this. This is also very high priority.
• It could use some optimizations. This is low priority.
• AlchemicalEliminationEngine and NCMCEngine - complete.
• Sampler classes (even if only using simple algorithms) need to be written in order to actually do anything with perses. This is high priority.

So, my near-term plan of action was, in order:

1. Finish relevant GeometryEngine stuff.
2. Fix the TopologyProposalEngine to correctly deal with multiple possible alignments.
3. Write a SAMSSampler and ExpandedEnsembleSampler.
4. Write an adaptive bias SAMSSampler.

After (3), we should be able to do relative free energy calculations with SAMS, as well as design with a fixed bias. I think that should be the initial goal here.

Does this sound ok?

[pgrinaway]

Ok, after a chat with @jchodera, we'll keep the above list except:

1. Basic API test framework
2. Finish relevant GeometryEngine stuff.
3. Fix the TopologyProposalEngine to correctly deal with multiple possible alignments.
4. Write a SAMSSampler, ExpandedEnsembleSampler, and adaptive bias SAMSSampler
5. Include examples (host/guest, T4 Lysozyme)

[jchodera]

Sounds great. We should include another example that tests @juliebehr's MutationEngine too. There are mutations of T4 lysozyme that modulate the ligand affinities: L99A binders, L99A nonbinders, L99A/M102Q binders, L99A/M102Q non-binders, and a number of additional mutants in this paper.

[pgrinaway]

Going to close this, since this is mostly done.