Framework example for non-star shaped analysis

[hannahbrucemacdonald]

I don't propose that this gets merged until it's in solid python, rather than in a notebook, but adding it here for version control/feedback.

This required changes to Arsenic, that I will open a PR for now

[hannahbrucemacdonald]

OpenFreeEnergy/cinnabar#17

To use this code, you will have to checkout the appropriate Arsenic branch until it is merged into master

[hannahbrucemacdonald]

Addressing #52

[hannahbrucemacdonald]

I'm not sure how best to return stuff from these functions, wether it should be the graph object itself (for plotting) or just a dictionary of {titles: (free_energy, estimate)}

[hannahbrucemacdonald]

TODO

• merge Arsenic changes that are required for this to work: Allowing for non-integer node names OpenFreeEnergy/cinnabar#17
• establish how best to pass the diffnet results around -- as nx.graphs, or populating the analysis json (probably both?)
• hold units all way throughout so that there can be no inconsistency

[glass-w]

I'm having trouble running the interactive networkx graph at:

for node in net.nodes:
    if node['title'] in has_expt:

Looks like node.keys() are: ['f_i', 'df_i', 'size', 'id', 'label', 'shape', 'font']

[hannahbrucemacdonald]

Hmmm I have dict_keys(['title', 'id', 'label', 'shape', 'font'])

What version of pyvis do you have?

[glass-w]

Using node['id'] seems to work instead, this is with networkx 2.4

[glass-w]

I think we'll want to return both the graph (for plotting purposes) and updated key entries to a .json for easy manipulation later.

[glass-w]

(or a ".json like" object e.g a list or dict)

[mcwitt]

Agreed that it would be useful to return both the graph and some json-compatible structure for inclusion in the analysis.json!

For the latter, I've been following the convention of describing the structure as a dataclass, e.g. like https://github.com/choderalab/covid-moonshot-infra/blob/266612d6ceef234ad6b77840f7e23cb65700dc64/covid_moonshot/core.py#L34-L40

This is nice as a form of documentation and also allows using something like mypy to check that we're not accessing fields incorrectly in downstream code.

For this case, would something like the following work?

@dataclass_json
@dataclass
class MoleculeAnalysis:
    molecule_id: str
    absolute_free_energy: FreeEnergy

def combine_relative_free_energies(results: Analysis) -> List[MoleculeAnalysis]:
    # ...

[glass-w]

this returns a float (?), need to update the docstring here

[mcwitt]

Nice, this works for me up until the last step, where I'm also having the issue @WG150 mentioned:

I'm having trouble running the interactive networkx graph at:

for node in net.nodes:
    if node['title'] in has_expt:

Looks like node.keys() are: ['f_i', 'df_i', 'size', 'id', 'label', 'shape', 'font']

I'm also seeing this with pyvis==0.1.8.2. When I replace title -> id the code runs, but the viz just shows a bunch of disconnected nodes. With title -> label, it runs and produces something that looks right, but without mouseover labels (@hannahbrucemacdonald, I thought I remembered your demo having these?)

[mcwitt]

I think we'll eventually want to transition this to taking a List[Run] or Analysis object rather than a dictionary (but this can be done later)

[mcwitt]

(I think we will want to merge #65 before doing this anyway)

[codecov-commenter]

Codecov Report

Merging #55 into master will increase coverage by 0.15%.
The diff coverage is 44.16%.