Model.wildtype_df BUG

[jgallowa07]

Was running through the spike analysis with multidms 1.9, and found a bug when using Model.wildtype_df.

It's predictions do not match what I get when I crunch the numbers directly from the parameters given in Model.get_mutations_df() + offsets. Concretely

Here is what the method is telling me.

             predicted_latent  predicted_func_score
Delta               -5.285840             -3.343554
Omicron_BA1         -0.853382             -0.103837
Omicron_BA2         -3.605787             -3.107373

but, when I predict the latent wildtype of BA2 by hand

converted_subs = model.data.convert_subs_wrt_ref_seq("Omicron_BA2", "")
print(converted_subs)
# 'T19I A27S V67A I95T I212L V213G L371F T376A D405N R408S S446G S496G K547T K856N F981L'
enc = bmap.sub_str_to_binary(converted_subs)
ba2_non_i_params = model.get_mutations_df(phenotype_as_effect=False).query("mutation.isin(@converted_subs.split())")
latent = (ba2_non_i_params.beta + ba2_non_i_params.shift_Omicron_BA2).sum()
offset = model.params["beta_naught"] + model.params["alpha_Omicron_BA2"]
BA2_wildtype = latent + offset
BA2_wildtype 
-0.7806519

returns $-0.7806519$ as opposed to $-3.605787$ as shown above.

This actually does not effect the spike analysis almost at all, however, that's because we're not looking at the WT "effect" in our analysis - but I'm certain this could be causing @jbloom issues (sorry! 😬).

That being said all the spike results identical! I'll be taking a look at this with fresh eyes tomorrow am.

[jgallowa07]

Er maybe it's not causing issues, yet, if you're only fitting single condition/experiment models

[jgallowa07]

Oh, I never pass the condition to from_latent or from_binary. Assuming this is the problem - will fix soon.

[jbloom]

I haven't noticed this yet because so far I'm just fitting a single condition.