Problem statement | John | Gromacs and OpenMM produce inconsistent results for binding free energies; any benchmarking we build would benefit from executing workloads for both We do want to use what’s already implemented via pmx if it is maintainable, but not clear without knowing what current workflow looks like Ideally, would be able to spin up a FAH work server on AWS whenever there is a new FF release to benchmark, preprocess PL library or job inputs, execute required simulations on FAH, then postprocess and ship data off to S3. After work is complete can shut down work server and hold on to EBS volume between boots
|
Scope
| John
| need system to execute both gromacs (via pmx currently) and openmm-based (perses) free energy evaluations bursty workload, so work server engineered for on-demand behavior pushes resulting data to S3; EBS volume preserved between boots local execution on work server for equilibration, pre- and post-processing can we use a cloudwatch trigger to boot up machine when a new ff spec is dropped into S3 input zone?
we also want any components we build for this to be usable for other execution efforts via FAH, so some generalizability is in scope, e.g. interested in BindingDB evaluation with same tooling Need to consider input mechanism PR based on e.g. PLBenchmarks would allow user submission, review process GHA could kick off work server to execute
Possible to nucleate around openmm_orion; need to assess how feasible this is; could be a win-win for us and OpenEye
|