Closed scottgigante-immunai closed 1 year ago
I think the right thing to do here is to construct an artificial anndata object for which we know exactly what the odds ratio should be and include that in the tests -- don't just test that it's a float but test that it returns a specific value.
Yep, agreed. Also, liana's rank_aggregate looks suspiciously different/better. Will also check that.
See https://deploy-preview-48--openproblems-sca.netlify.app/benchmarks/cell_cell_communication_ligand_target/tnbc_data/