Sounds good. I'll start thinking about general code to implement this. Are there more complex examples that need to be handled than

['OR', ['AND', 'A', 'B'], ['AND', 'C', 'D']]
?

It is certainly possible to imagine something with more than two stereo groups. One problem with working on this is that there are not a lot of files out there that have real-world examples of molecules with enhanced stereo (at least not that I'm aware of).

Thanks. Re-reading your blog post, I see another example:

And it would also ideally be able to recognize that: ['AND', ['OR', 'A', 'B'], ['OR', 'C', 'D']] is:

  'A' AND 'C'
OR
  'A' AND 'D'
OR
  'B' AND 'C'
OR
  'B' AND 'D'

Also I recall @teherin (if I have the correct Erin Geno, from @FlareTx) showing some examples of enhanced and empathetic 😀 stereochemistry at the North American RDKit UGM on April 10. If Erin could share those, that would be helpful.

You've got the right me. I've also been using MolsMatrixToGridImage() as a convenient way to represent the possible ensembles, I think it's a great use.
When enumerating ES presenting it to chemists as "here's the mixture (or options for mixtures) in your flask" has always resonated. This is where I think the list of lists concept for depiction shines.

@marcostenta I'd be curious to see how med chemists react to a tree structure, my gut instinct is some sizeable minority would get pretty confused. I might be wrong, in either case I could see it as a useful teaching tool. Also you would have to make some decisions about how to depict the branching. For example: do you start with ands, ors?

Examples wise the only vendor I recall that uses it ES consistently is Enamine in their BB catalogs, but those are typically 1 or 2 enhanced centers max.
Here's a small set of some hypothetical examples based on simplified Suzetrigine. I have a feeling we may need to generate our own test sets.
es_analogs = {"cis_one_unk": "C[C@@H]1[C@@H]([C@@H](O[C@@]1(C)C(F)(F)F)C(=O)Nc1ccncc1)c1ccccc1 |a:5,o1:3,&1:1,2|", "trans_one_unk": "C[C@@H]1[C@H]([C@@H](O[C@@]1(C)C(F)(F)F)C(=O)Nc1ccncc1)c1ccccc1 |a:5,o1:3,&1:1,2|", "cis_two_unk": "C[C@@H]1[C@@H]([C@@H](O[C@@]1(C)C(F)(F)F)C(=O)Nc1ccncc1)c1ccccc1 |o1:3,o2:5,&1:1,2|", "rac_3_unk": "CC1[C@H]([C@@H](O[C@@]1(C)C(F)(F)F)C(=O)Nc1ccncc1)c1ccccc1 |o1:3,o2:5,o3:2|"}

@teherin med chems would react pretty bad to a tree structure in general but it can serve some chemoinformatics and didactic purposes (and also good for testing). From a tree structure we could derive the replationship between the isomers. I got the idea from a slide-deck from chemaxon https://www.slideshare.net/slideshow/enhanced-stereochemistry-representation/247244647 (slide 28). In this case they define the tree as a way to represent the separation process of a mixture, but the same tree could represent (at least internally) the logic tree.
I agree about the test set: we also need a signature of sort to use in the test assert stements, that is unambiguous and easy to generate manually (to compare with the compute one)