Return 0 for patients with no neoantigens

[arahuja]

Removed the check of IDs since this give NaN for patients without neoantigens

[coveralls]

Coverage increased (+0.05%) to 57.87% when pulling 218e077 on neoantigen-nan into 7aed533 on master.

[coveralls]

Coverage decreased (-0.8%) to 57.031% when pulling 218e077 on neoantigen-nan into 7aed533 on master.

[tavinathanson]

That was intended, and downstream the code checks for null and if null prints that patient X was skipped; which seems important when e.g. cohort = data.init_cohort(only_patients_with_bams=False) (sorry to reference a private repo here).

At a high level, it seems useful/necessary to maintain a distinction between 0 and not available? Where is the NaN a problem for you?

[arahuja]

@tavinathanson in this case the correct answer is 0 though and not NaN (filtering a dataframe for an ID when there are 0 rows that match should be 0)

Compared to the other cases where we are checking if there is an entry corresponding to the ID and then using len. Here there are actually 0 neoantigens so there are 0 rows in the dataframe (and the ID is also not in the dataframe). Alternatively, we can add a separate check that we ran the neoantigens part for this ID.

Also, I haven't found this useful for the mutation counts either, since if a patient is missing VCF files, we return an empty VariantCollection. Which means the check for the ID passes, and we (falsely) return 0 as opposed to NaN

[tavinathanson]

Ah, I see. I guess we could look at what mutations are present to decide between 0 and NaN?

Re the empty VariantCollection: great point. It used to filter them out properly, so that's a bug.

[tavinathanson]

Merge away per offline discussion:

@tavinathanson: still not clear to me where you're running into issues with nan?
@arahuja: When there are no neoantigens for example, plot_benefit etc throw out some of the patients
@tavinathanson: suppose you can merge it in and i can follow it up with a PR that brings back nan in a non buggy way, sound reasonable?
@arahuja: Sure - what were you thinking though?
@tavinathanson: using load_variants to distinguish 0 from not available; and for not available, we should throw em out (not fully formed; does that make sense?)
@arahuja: yea but it could be variants were available but hla wasn't
@tavinathanson: can we rely on whether or not the neoantigen cache folder for that patient exists?
@arahuja: not sure when it’s created (i.e. whether it’s created regardless and then just populated if anything is missing)
@tavinathanson: i think not created regardless but have to check