Hi Anna,

Currently, the extend_formula_ncem function (referenced below) computes a list of coef_couplings:

coef_couplings = [f"{PREFIX_INDEX}{x}:{PREFIX_NEIGHBOR}{y}" for y in cell_types for x in cell_types]

For my cell types, this list looks like:

['index_AcinarCells:neighbor_AcinarCells', 'index_AcinarCells:neighbor_B', 'index_AcinarCells:neighbor_Basal', ...]

However, the get_dmats_from_deconvoluted generates a different list of coefficients. When I inspect dmats[x].design_info.column_names, I instead see

['index_AcinarCells[False]:neighbor_AcinarCells', 'index_AcinarCells[True]:neighbor_AcinarCells', 'index_AcinarCells[False]:neighbor_B', 'index_AcinarCells[True]:neighbor_B', ...]

Note the addition of [False] and [True]. The fact that these lists are different leads to problems when we run test_deconvoluted.

I think this issue can be fixed by updating the extend_formula_ncem function. I'll take a look at it and let you know when I have a solution.